diff --git a/config/op_info.config b/config/op_info.config
index 44db8f63dfc..ca69228fe21 100644
--- a/config/op_info.config
+++ b/config/op_info.config
@@ -14,6 +14,7 @@
 {"op_name": "DropoutGenMask", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [{"name": "Seed0", "type": "int"}, {"name": "Seed1", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Dropout2D", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}, {"index": 1, "name": "mask", "param_type": "required"}], "attr": [{"name": "keep_prob", "type": "float"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Dropout3D", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}, {"index": 1, "name": "mask", "param_type": "required"}], "attr": [{"name": "keep_prob", "type": "float"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
+{"op_name": "DynamicStitch", "inputs": [{"index": 0, "name": "indices", "param_type": "dynamic"}, {"index": 1, "name": "data", "param_type": "dynamic"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "GetNext", "inputs": [], "outputs": [{"index": 0, "name": "y", "param_type": "dynamic"}], "attr": [{"name": "shared_name", "type": "str"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"]], [["int8", "DefaultFormat"]], [["int16", "DefaultFormat"]], [["int32", "DefaultFormat"]], [["int64", "DefaultFormat"]], [["float16", "DefaultFormat"]], [["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Print", "inputs": [{"index": 0, "name": "x", "param_type": "dynamic"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "TopK", "inputs": [{"index": 0, "name": "input", "param_type": "required"}, {"index": 1, "name": "k", "param_type": "required"}], "outputs": [{"index": 0, "name": "values", "param_type": "required"}, {"index": 1, "name": "indices", "param_type": "required"}], "attr": [{"name": "sorted", "type": "bool"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "AiCPU"}
@@ -24,6 +25,7 @@
 {"op_name": "ExpandDims", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Randperm", "inputs": [{"index": 0, "name": "n", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [{"name": "max_length", "type": "int"}, {"name": "pad", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "RandomChoiceWithMask", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}, {"index": 1, "name": "mask", "param_type": "required"}], "attr": [{"name": "count", "type": "int"}, {"name": "seed", "type": "int"}, {"name": "seed2", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
+{"op_name": "SearchSorted", "inputs": [{"index": 0, "name": "sequence", "param_type": "required"}, {"index": 1, "name": "values", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [{"name": "out_int32", "type": "bool"}, {"name": "right", "type": "bool"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Stack", "inputs": [{"index": 0, "name": "x", "param_type": "dynamic"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [{"name": "axis", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "UniformCandidateSampler", "inputs": [{"index": 0, "name": "true_classes", "param_type": "required"}], "outputs": [{"index": 0, "name": "sampled_candidates", "param_type": "required"}, {"index": 1, "name": "true_expected_count", "param_type": "required"}, {"index": 2, "name": "true_expected_count", "param_type": "required"}], "attr": [{"name": "num_true", "type": "int"}, {"name": "num_sampled", "type": "int"}, {"name": "unique", "type": "bool"}, {"name": "range_max", "type": "int"}, {"name": "seed", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "LogUniformCandidateSampler", "inputs": [{"index": 0, "name": "true_classes", "param_type": "required"}], "outputs": [{"index": 0, "name": "sampled_candidates", "param_type": "required"}, {"index": 1, "name": "true_expected_count", "param_type": "required"}, {"index": 2, "name": "true_expected_count", "param_type": "required"}], "attr": [{"name": "num_true", "type": "int"}, {"name": "num_sampled", "type": "int"}, {"name": "unique", "type": "bool"}, {"name": "range_max", "type": "int"}, {"name": "seed", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
@@ -35,6 +37,8 @@
 {"op_name": "RandomCategorical", "inputs": [{"index": 0, "name": "logits", "param_type": "required"}, {"index": 1, "name": "num_sample", "param_type": "required"}, {"index": 2, "name": "seed", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Cast", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "MirrorPad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "paddings", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [{"name": "mode", "type": "str"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "AiCPU"}
+{"op_name": "MaskedSelect", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "mask", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["bool", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
+{"op_name": "MaskedSelectGrad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "mask", "param_type": "required"}, {"index": 2, "name": "grad", "param_type": "required"}], "outputs": [{"index": 0, "name": "dx", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "MirrorPadGrad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "paddings", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [{"name": "mode", "type": "str"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "StandardNormal", "inputs": [{"index": 0, "name": "shape", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [{"name": "seed", "type": "int"}, {"name": "seed2", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "Gamma", "inputs": [{"index": 0, "name": "shape", "param_type": "required"}, {"index": 1, "name": "alpha", "param_type": "required"}, {"index": 2, "name": "beta", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [{"name": "seed", "type": "int"}, {"name": "seed2", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "AiCPU"}
@@ -58,6 +62,54 @@
 {"op_name": "StackPop", "inputs": [], "outputs": [{"index": 0, "name": "dst", "param_type": "required"}], "attr": [{"name": "index", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [[["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"]], [["int8", "DefaultFormat"]], [["int16", "DefaultFormat"]], [["int32", "DefaultFormat"]], [["int64", "DefaultFormat"]], [["float16", "DefaultFormat"]], [["float32", "DefaultFormat"]], [["float64", "DefaultFormat"]], [["bool", "DefaultFormat"]]], "imply_type": "AiCPU"}
 {"op_name": "StackDestroy", "inputs": [], "outputs": [], "attr": [{"name": "index", "type": "int"}], "fusion_type": "OPAQUE", "dtype_format": [], "imply_type": "AiCPU"}
 {"op_name": "CTCGreedyDecoder", "inputs": [{"index": 0, "name": "inputs", "param_type": "required"}, {"index": 1, "name": "sequence_length", "param_type": "required"}], "outputs": [{"index": 0, "name": "decoded_indices", "param_type": "required"}, {"index": 1, "name": "decoded_values", "param_type": "required"}, {"index": 2, "name": "decoded_shape", "param_type": "required"}, {"index": 3, "name": "log_probability", "param_type": "required"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "AiCPU"}
+{"op_name": "Cast", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float16", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["float64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Mul", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Sub", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Pow", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "RealDiv", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Div", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Concat", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Split", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Adam", "inputs": [{"index": 0, "name": "var", "param_type": "required"}, {"index": 1, "name": "m", "param_type": "required"}, {"index": 2, "name": "v", "param_type": "required"}, {"index": 3, "name": "beta1_power", "param_type": "required"}, {"index": 4, "name": "beta2_power", "param_type": "required"}, {"index": 5, "name": "lr", "param_type": "required"}, {"index": 6, "name": "beta1", "param_type": "required"}, {"index": 7, "name": "beta2", "param_type": "required"}, {"index": 8, "name": "epsilon", "param_type": "required"}, {"index": 9, "name": "gradient", "param_type": "required"}], "outputs": [{"index": 0, "name": "output0", "param_type": "required"}, {"index": 1, "name": "output0", "param_type": "required"}, {"index": 2, "name": "output0", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Argmax", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ArgMinWithValue", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "indice", "param_type": "required"}, {"index": 1, "name": "values", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "BiasAdd", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "bias", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "BiasAddGrad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Dropout", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "output0", "param_type": "required"}, {"index": 1, "name": "output1", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "DropoutGrad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "y", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "GatherD", "inputs": [{"index": 0, "name": "input", "param_type": "required"}, {"index": 1, "name": "dim", "param_type": "required"}, {"index": 2, "name": "index", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "GatherDGrad", "inputs": [{"index": 0, "name": "index", "param_type": "required"}, {"index": 1, "name": "src", "param_type": "required"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Gather", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "indices", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float64", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "GatherNd", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Maximum", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "MaximumGrad", "inputs": [{"index": 0, "name": "grads", "param_type": "required"}, {"index": 1, "name": "x1", "param_type": "required"}, {"index": 2, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y1", "param_type": "required"}, {"index": 1, "name": "y2", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Conv2D", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "filter", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "HSigmoid", "inputs": [{"index": 0, "name": "x"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "HSigmoidGrad", "inputs": [{"index": 0, "name": "y_grad"}, {"index": 1, "name": "x"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "HSwish", "inputs": [{"index": 0, "name": "x"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "HSwishGrad", "inputs": [{"index": 0, "name": "y_grad"}, {"index": 1, "name": "x"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "IsFinite", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float64", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "LayerNorm", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "gamma", "param_type": "required"}, {"index": 2, "name": "beta", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}, {"index": 1, "name": "mean", "param_type": "required"}, {"index": 2, "name": "variance", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "LayerNormGrad", "inputs": [{"index": 0, "name": "dy", "param_type": "required"}, {"index": 1, "name": "x", "param_type": "required"}, {"index": 2, "name": "variance", "param_type": "required"}, {"index": 3, "name": "mean", "param_type": "required"}, {"index": 4, "name": "gamma", "param_type": "required"}], "outputs": [{"index": 0, "name": "pd_x", "param_type": "required"}, {"index": 1, "name": "pd_gamma", "param_type": "required"}, {"index": 2, "name": "pd_beta", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Minimum", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "MinimumGrad", "inputs": [{"index": 0, "name": "grads", "param_type": "required"}, {"index": 1, "name": "x1", "param_type": "required"}, {"index": 2, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y1", "param_type": "required"}, {"index": 1, "name": "y2", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"], ["float64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "EqualCount", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "MirrorPad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "paddings", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "MirrorPadGrad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "paddings", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Stack", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceMean", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceMax", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceSum", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceMin", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float64", "DefaultFormat"], ["float64", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceAll", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "ReduceAny", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Transpose", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Tile", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "TopK", "inputs": [{"index": 0, "name": "input", "param_type": "required"}, {"index": 1, "name": "k", "param_type": "required"}], "outputs": [{"index": 0, "name": "values", "param_type": "required"}, {"index": 1, "name": "indices", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Add", "inputs": [{"index": 0, "name": "x1", "param_type": "required"}, {"index": 1, "name": "x2", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "OneHot", "inputs": [{"index": 0, "name": "x", "param_type": "required"}, {"index": 1, "name": "on_value", "param_type": "required"}, {"index": 2, "name": "off_value", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Pad", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "CPU"}
+{"op_name": "Range", "inputs": [{"index": 0, "name": "start", "param_type": "required"}, {"index": 1, "name": "limit"}, {"index": 2, "name": "delta"}], "outputs": [{"index": 0, "name": "output", "param_type": "required"}], "attr": [], "fusion_type": "", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "CPU"}
 {"op_name": "Abs", "inputs": [{"index": 0, "name": "x"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]]], "imply_type": "AKG", "processor": "AiCore"}
 {"op_name": "Add", "inputs": [{"index": 0, "name": "x"}, {"index": 1, "name": "y"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"], ["int32", "NC1HWC0"]], [["float16", "FracZ"], ["float16", "FracZ"], ["float16", "FracZ"]], [["float32", "FracZ"], ["float32", "FracZ"], ["float32", "FracZ"]], [["int32", "FracZ"], ["int32", "FracZ"], ["int32", "FracZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["int32", "FRACTAL_NZ"], ["int32", "FRACTAL_NZ"], ["int32", "FRACTAL_NZ"]]], "imply_type": "AKG", "processor": "AiCore"}
 {"op_name": "AddN", "inputs": [{"index": 0, "name": "inputs", "param_type": "dynamic"}], "outputs": [{"index": 0, "name": "output"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float16", "FracZ"], ["float16", "FracZ"]], [["float32", "FracZ"], ["float32", "FracZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]]], "imply_type": "AKG", "processor": "AiCore"}
@@ -141,6 +193,7 @@
 {"op_name": "Conv2DBackpropInput", "inputs": [{"index": 0, "name": "out_backprop", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "input_sizes", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "stride", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_list", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "dilation", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "groups", "param_type": "optional", "type": "int", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "CONVLUTION", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "FracZ"], ["float16", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "conv2d_backprop_input_d.so", "compute_cost": 10, "kernel_name": "conv2d_backprop_input_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ConfusionMulGrad", "inputs": [{"index": 0, "name": "input0", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output0", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "required", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "confusion_mul_grad.so", "compute_cost": 10, "kernel_name": "confusion_mul_grad", "partial_flag": false, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "DropoutDoMask", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "keep_prob", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""], ["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "drop_out_do_mask.so", "compute_cost": 10, "kernel_name": "drop_out_do_mask", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
+{"op_name": "DropoutDoMask", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "keep_prob", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""], ["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "drop_out_do_mask.so", "compute_cost": 10, "kernel_name": "drop_out_do_mask", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "GeLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gelu.so", "compute_cost": 10, "kernel_name": "gelu", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "GeLUGrad", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "z", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gelu_grad.so", "compute_cost": 10, "kernel_name": "gelu_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "FastGeLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fast_gelu.so", "compute_cost": 10, "kernel_name": "fast_gelu", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
@@ -159,23 +212,27 @@
 {"op_name": "ReLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu.so", "compute_cost": 10, "kernel_name": "relu", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "ReLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu.so", "compute_cost": 10, "kernel_name": "relu", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "ReluGrad", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "features", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""], ["uint8", ""]], [["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_grad.so", "compute_cost": 10, "kernel_name": "relu_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
+{"op_name": "ReluGrad", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "features", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""], ["uint8", ""]], [["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_grad.so", "compute_cost": 10, "kernel_name": "relu_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "ReLU6", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]], [["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu6.so", "compute_cost": 10, "kernel_name": "relu6", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "ReLU6Grad", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "features", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu6_grad.so", "compute_cost": 10, "kernel_name": "relu6_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ReLUV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["int8", "NC1HWC0"], ["int8", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["uint8", "NC1HWC0"], ["uint8", "NC1HWC0"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_v2.so", "compute_cost": 10, "kernel_name": "relu_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ReluGradV2", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["uint8", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["uint8", "DefaultFormat"], ["float32", "NC1HWC0"]], [["int32", "NC1HWC0"], ["uint8", "DefaultFormat"], ["int32", "NC1HWC0"]], [["int8", "NC1HWC0"], ["uint8", "DefaultFormat"], ["int8", "NC1HWC0"]], [["uint8", "NC1HWC0"], ["uint8", "DefaultFormat"], ["uint8", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_grad_v2.so", "compute_cost": 10, "kernel_name": "relu_grad_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "ReLUV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["int8", "NC1HWC0"], ["int8", "NC1HWC0"], ["uint8", "DefaultFormat"]], [["uint8", "NC1HWC0"], ["uint8", "NC1HWC0"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_v2.so", "compute_cost": 10, "kernel_name": "relu_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "ReluGradV2", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mask", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["uint8", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["uint8", "DefaultFormat"], ["float32", "NC1HWC0"]], [["int32", "NC1HWC0"], ["uint8", "DefaultFormat"], ["int32", "NC1HWC0"]], [["int8", "NC1HWC0"], ["uint8", "DefaultFormat"], ["int8", "NC1HWC0"]], [["uint8", "NC1HWC0"], ["uint8", "DefaultFormat"], ["uint8", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "relu_grad_v2.so", "compute_cost": 10, "kernel_name": "relu_grad_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SoftmaxCrossEntropyWithLogits", "inputs": [{"index": 0, "name": "input_features", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input_labels", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output_loss", "need_compile": true, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output_backprop", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softmax_cross_entropy_with_logits.so", "compute_cost": 10, "kernel_name": "softmax_cross_entropy_with_logits", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SigmoidCrossEntropyWithLogits", "inputs": [{"index": 0, "name": "predict", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "loss", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NDC1HWC0"], ["float32", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sigmoid_cross_entropy_with_logits.so", "compute_cost": 10, "kernel_name": "sigmoid_cross_entropy_with_logits", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SigmoidCrossEntropyWithLogitsGrad", "inputs": [{"index": 0, "name": "predict", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dout", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "gradient", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NDC1HWC0"], ["float32", "NDC1HWC0"], ["float32", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sigmoid_cross_entropy_with_logits_grad.so", "compute_cost": 10, "kernel_name": "sigmoid_cross_entropy_with_logits_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Add", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"], ["int32", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "add.so", "compute_cost": 10, "kernel_name": "add", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "Add", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "NC1HWC0"], ["int32", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "add.so", "compute_cost": 10, "kernel_name": "add", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "TransData", "inputs": [{"index": 0, "name": "src", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dst", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "src_format", "param_type": "required", "type": "str", "value": "DefaultFormat, NC1HWC0, FracZ, FRACTAL_NZ, HWCN, C1HWNCoC0, NDHWC, NHWC"}, {"name": "dst_format", "param_type": "required", "type": "str", "value": "DefaultFormat, NC1HWC0, FracZ, FRACTAL_NZ, HWCN, C1HWNCoC0, NDHWC, NHWC"}, {"name": "groups", "param_type": "optional", "type": "int", "value": "all", "default_value": "1"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "NHWC"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NHWC"]], [["float32", "NC1HWC0"], ["float32", "DefaultFormat"]], [["float32", "FracZ"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "FracZ"]], [["float32", "HWCN"], ["float32", "FracZ"]], [["float32", "FracZ"], ["float32", "HWCN"]], [["float32", "C1HWNCoC0"], ["float32", "HWCN"]], [["float32", "HWCN"], ["float32", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "NHWC"], ["float16", "FracZ"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float16", "NHWC"], ["float16", "NC1HWC0"]], [["float16", "HWCN"], ["float16", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NHWC"]], [["float16", "NC1HWC0"], ["float16", "DefaultFormat"]], [["float16", "FracZ"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "FracZ"], ["float16", "HWCN"]], [["float16", "C1HWNCoC0"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["float16", "FRACTAL_NZ"]], [["float32", "DefaultFormat"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "DefaultFormat"]], [["float32", "FRACTAL_NZ"], ["float32", "DefaultFormat"]], [["bool", "NHWC"], ["bool", "NC1HWC0"]], [["bool", "DefaultFormat"], ["bool", "NC1HWC0"]], [["bool", "NC1HWC0"], ["bool", "NHWC"]], [["bool", "NC1HWC0"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "NHWC"]], [["float16", "DefaultFormat"], ["float16", "HWCN"]], [["float16", "NHWC"], ["float16", "DefaultFormat"]], [["float16", "NHWC"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "DefaultFormat"]], [["float16", "HWCN"], ["float16", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "HWCN"]], [["float32", "NHWC"], ["float32", "DefaultFormat"]], [["float32", "NHWC"], ["float32", "HWCN"]], [["float32", "HWCN"], ["float32", "DefaultFormat"]], [["float32", "HWCN"], ["float32", "NHWC"]], [["int8", "DefaultFormat"], ["int8", "FRACTAL_NZ"]], [["int8", "DefaultFormat"], ["int8", "FracZ"]], [["int8", "DefaultFormat"], ["int8", "NHWC"]], [["int8", "DefaultFormat"], ["int8", "HWCN"]], [["int8", "NHWC"], ["int8", "DefaultFormat"]], [["int8", "NHWC"], ["int8", "HWCN"]], [["int8", "HWCN"], ["int8", "DefaultFormat"]], [["int8", "HWCN"], ["int8", "NHWC"]], [["int16", "DefaultFormat"], ["int16", "NHWC"]], [["int16", "DefaultFormat"], ["int16", "HWCN"]], [["int16", "NHWC"], ["int16", "DefaultFormat"]], [["int16", "NHWC"], ["int16", "HWCN"]], [["int16", "HWCN"], ["int16", "DefaultFormat"]], [["int16", "HWCN"], ["int16", "NHWC"]], [["int32", "DefaultFormat"], ["int32", "NHWC"]], [["int32", "DefaultFormat"], ["int32", "HWCN"]], [["int32", "NHWC"], ["int32", "DefaultFormat"]], [["int32", "NHWC"], ["int32", "HWCN"]], [["int32", "HWCN"], ["int32", "DefaultFormat"]], [["int32", "HWCN"], ["int32", "NHWC"]], [["int64", "DefaultFormat"], ["int64", "NHWC"]], [["int64", "DefaultFormat"], ["int64", "HWCN"]], [["int64", "NHWC"], ["int64", "DefaultFormat"]], [["int64", "NHWC"], ["int64", "HWCN"]], [["int64", "HWCN"], ["int64", "DefaultFormat"]], [["int64", "HWCN"], ["int64", "NHWC"]], [["uint8", "DefaultFormat"], ["uint8", "NHWC"]], [["uint8", "DefaultFormat"], ["uint8", "HWCN"]], [["uint8", "NHWC"], ["uint8", "DefaultFormat"]], [["uint8", "NHWC"], ["uint8", "HWCN"]], [["uint8", "HWCN"], ["uint8", "DefaultFormat"]], [["uint8", "HWCN"], ["uint8", "NHWC"]], [["uint16", "DefaultFormat"], ["uint16", "NHWC"]], [["uint16", "DefaultFormat"], ["uint16", "HWCN"]], [["uint16", "NHWC"], ["uint16", "DefaultFormat"]], [["uint16", "NHWC"], ["uint16", "HWCN"]], [["uint16", "HWCN"], ["uint16", "DefaultFormat"]], [["uint16", "HWCN"], ["uint16", "NHWC"]], [["uint32", "DefaultFormat"], ["uint32", "NHWC"]], [["uint32", "DefaultFormat"], ["uint32", "HWCN"]], [["uint32", "NHWC"], ["uint32", "DefaultFormat"]], [["uint32", "NHWC"], ["uint32", "HWCN"]], [["uint32", "HWCN"], ["uint32", "DefaultFormat"]], [["uint32", "HWCN"], ["uint32", "NHWC"]], [["uint64", "DefaultFormat"], ["uint64", "NHWC"]], [["uint64", "DefaultFormat"], ["uint64", "HWCN"]], [["uint64", "NHWC"], ["uint64", "DefaultFormat"]], [["uint64", "NHWC"], ["uint64", "HWCN"]], [["uint64", "HWCN"], ["uint64", "DefaultFormat"]], [["uint64", "HWCN"], ["uint64", "NHWC"]], [["int32", "FRACTAL_NZ"], ["int32", "DefaultFormat"]], [["float16", "NDHWC"], ["float16", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NDHWC"]], [["int8", "HWCN"], ["int8", "C1HWNCoC0"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "FracZ"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "FRACTAL_NZ"]], [["float32", "HWCN"], ["float16", "FRACTAL_NZ"]], [["float16", "HWCN"], ["float16", "FRACTAL_ZN_LSTM"]], [["float32", "HWCN"], ["float32", "FRACTAL_ZN_LSTM"]], [["float16", "FRACTAL_ZN_LSTM"], ["float16", "HWCN"]], [["float32", "FRACTAL_ZN_LSTM"], ["float32", "HWCN"]], [["float16", "NDHWC"], ["float16", "NDC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NDHWC"]], [["float16", "DHWCN"], ["float16", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "DHWCN"]], [["float16", "NCDHW"], ["float16", "NDC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NCDHW"]], [["float16", "NCDHW"], ["float16", "FRACTAL_Z_3D"]], [["float32", "NCDHW"], ["float32", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "NCDHW"]], [["float32", "FRACTAL_Z_3D"], ["float32", "NCDHW"]], [["float16", "NDHWC"], ["float16", "FRACTAL_Z_3D"]], [["float32", "NDHWC"], ["float32", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "NDHWC"]], [["float32", "FRACTAL_Z_3D"], ["float32", "NDHWC"]], [["float32", "DHWCN"], ["float32", "FRACTAL_Z_3D"]], [["float32", "FRACTAL_Z_3D"], ["float32", "DHWCN"]], [["float32", "NDC1HWC0"], ["float32", "NDHWC"]], [["float32", "NDHWC"], ["float32", "NDC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NCDHW"]], [["float32", "NCDHW"], ["float32", "NDC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NCDHW"]], [["float32", "NCDHW"], ["float32", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "trans_data.so", "compute_cost": 10, "kernel_name": "trans_data", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "TopK", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "assist_seq", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "values", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "dim", "param_type": "optional", "type": "int", "value": "all"}, {"name": "k", "param_type": "required", "type": "int", "value": "all"}, {"name": "largest", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "sorted", "param_type": "optional", "type": "bool", "value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "top_k_d.so", "compute_cost": 10, "kernel_name": "top_k_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "TransData", "inputs": [{"index": 0, "name": "src", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dst", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "src_format", "param_type": "required", "type": "str", "value": "DefaultFormat, NC1HWC0, FracZ, FRACTAL_NZ, HWCN, C1HWNCoC0, NDHWC, NHWC"}, {"name": "dst_format", "param_type": "required", "type": "str", "value": "DefaultFormat, NC1HWC0, FracZ, FRACTAL_NZ, HWCN, C1HWNCoC0, NDHWC, NHWC"}, {"name": "groups", "param_type": "optional", "type": "int", "value": "all", "default_value": "1"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "NHWC"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NHWC"]], [["float32", "NC1HWC0"], ["float32", "DefaultFormat"]], [["float32", "FracZ"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "FracZ"]], [["float32", "HWCN"], ["float32", "FracZ"]], [["float32", "FracZ"], ["float32", "HWCN"]], [["float32", "C1HWNCoC0"], ["float32", "HWCN"]], [["float32", "HWCN"], ["float32", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "NHWC"], ["float16", "FracZ"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float16", "NHWC"], ["float16", "NC1HWC0"]], [["float16", "HWCN"], ["float16", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NHWC"]], [["float16", "NC1HWC0"], ["float16", "DefaultFormat"]], [["float16", "FracZ"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "FracZ"], ["float16", "HWCN"]], [["float16", "C1HWNCoC0"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["float16", "FRACTAL_NZ"]], [["float32", "DefaultFormat"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "DefaultFormat"]], [["float32", "FRACTAL_NZ"], ["float32", "DefaultFormat"]], [["bool", "NHWC"], ["bool", "NC1HWC0"]], [["bool", "DefaultFormat"], ["bool", "NC1HWC0"]], [["bool", "NC1HWC0"], ["bool", "NHWC"]], [["bool", "NC1HWC0"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "NHWC"]], [["float16", "DefaultFormat"], ["float16", "HWCN"]], [["float16", "NHWC"], ["float16", "DefaultFormat"]], [["float16", "NHWC"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "DefaultFormat"]], [["float16", "HWCN"], ["float16", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "HWCN"]], [["float32", "NHWC"], ["float32", "DefaultFormat"]], [["float32", "NHWC"], ["float32", "HWCN"]], [["float32", "HWCN"], ["float32", "DefaultFormat"]], [["float32", "HWCN"], ["float32", "NHWC"]], [["int8", "DefaultFormat"], ["int8", "FRACTAL_NZ"]], [["int8", "DefaultFormat"], ["int8", "FracZ"]], [["int8", "DefaultFormat"], ["int8", "NHWC"]], [["int8", "DefaultFormat"], ["int8", "HWCN"]], [["int8", "NHWC"], ["int8", "DefaultFormat"]], [["int8", "NHWC"], ["int8", "HWCN"]], [["int8", "HWCN"], ["int8", "DefaultFormat"]], [["int8", "HWCN"], ["int8", "NHWC"]], [["int16", "DefaultFormat"], ["int16", "NHWC"]], [["int16", "DefaultFormat"], ["int16", "HWCN"]], [["int16", "NHWC"], ["int16", "DefaultFormat"]], [["int16", "NHWC"], ["int16", "HWCN"]], [["int16", "HWCN"], ["int16", "DefaultFormat"]], [["int16", "HWCN"], ["int16", "NHWC"]], [["int32", "DefaultFormat"], ["int32", "NHWC"]], [["int32", "DefaultFormat"], ["int32", "HWCN"]], [["int32", "NHWC"], ["int32", "DefaultFormat"]], [["int32", "NHWC"], ["int32", "HWCN"]], [["int32", "HWCN"], ["int32", "DefaultFormat"]], [["int32", "HWCN"], ["int32", "NHWC"]], [["int64", "DefaultFormat"], ["int64", "NHWC"]], [["int64", "DefaultFormat"], ["int64", "HWCN"]], [["int64", "NHWC"], ["int64", "DefaultFormat"]], [["int64", "NHWC"], ["int64", "HWCN"]], [["int64", "HWCN"], ["int64", "DefaultFormat"]], [["int64", "HWCN"], ["int64", "NHWC"]], [["uint8", "DefaultFormat"], ["uint8", "NHWC"]], [["uint8", "DefaultFormat"], ["uint8", "HWCN"]], [["uint8", "NHWC"], ["uint8", "DefaultFormat"]], [["uint8", "NHWC"], ["uint8", "HWCN"]], [["uint8", "HWCN"], ["uint8", "DefaultFormat"]], [["uint8", "HWCN"], ["uint8", "NHWC"]], [["uint16", "DefaultFormat"], ["uint16", "NHWC"]], [["uint16", "DefaultFormat"], ["uint16", "HWCN"]], [["uint16", "NHWC"], ["uint16", "DefaultFormat"]], [["uint16", "NHWC"], ["uint16", "HWCN"]], [["uint16", "HWCN"], ["uint16", "DefaultFormat"]], [["uint16", "HWCN"], ["uint16", "NHWC"]], [["uint32", "DefaultFormat"], ["uint32", "NHWC"]], [["uint32", "DefaultFormat"], ["uint32", "HWCN"]], [["uint32", "NHWC"], ["uint32", "DefaultFormat"]], [["uint32", "NHWC"], ["uint32", "HWCN"]], [["uint32", "HWCN"], ["uint32", "DefaultFormat"]], [["uint32", "HWCN"], ["uint32", "NHWC"]], [["uint64", "DefaultFormat"], ["uint64", "NHWC"]], [["uint64", "DefaultFormat"], ["uint64", "HWCN"]], [["uint64", "NHWC"], ["uint64", "DefaultFormat"]], [["uint64", "NHWC"], ["uint64", "HWCN"]], [["uint64", "HWCN"], ["uint64", "DefaultFormat"]], [["uint64", "HWCN"], ["uint64", "NHWC"]], [["int32", "FRACTAL_NZ"], ["int32", "DefaultFormat"]], [["float16", "NDHWC"], ["float16", "NC1HWC0"]], [["float16", "NC1HWC0"], ["float16", "NDHWC"]], [["int8", "HWCN"], ["int8", "C1HWNCoC0"]], [["float16", "HWCN"], ["float16", "FracZ"]], [["float16", "FracZ"], ["float16", "HWCN"]], [["float16", "HWCN"], ["float16", "FRACTAL_NZ"]], [["float32", "HWCN"], ["float16", "FRACTAL_NZ"]], [["float16", "HWCN"], ["float16", "FRACTAL_ZN_LSTM"]], [["float32", "HWCN"], ["float32", "FRACTAL_ZN_LSTM"]], [["float16", "FRACTAL_ZN_LSTM"], ["float16", "HWCN"]], [["float32", "FRACTAL_ZN_LSTM"], ["float32", "HWCN"]], [["float16", "NDHWC"], ["float16", "NDC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NDHWC"]], [["float16", "DHWCN"], ["float16", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "DHWCN"]], [["float16", "NCDHW"], ["float16", "NDC1HWC0"]], [["float16", "NDC1HWC0"], ["float16", "NCDHW"]], [["float16", "NCDHW"], ["float16", "FRACTAL_Z_3D"]], [["float32", "NCDHW"], ["float32", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "NCDHW"]], [["float32", "FRACTAL_Z_3D"], ["float32", "NCDHW"]], [["float16", "NDHWC"], ["float16", "FRACTAL_Z_3D"]], [["float32", "NDHWC"], ["float32", "FRACTAL_Z_3D"]], [["float16", "FRACTAL_Z_3D"], ["float16", "NDHWC"]], [["float32", "FRACTAL_Z_3D"], ["float32", "NDHWC"]], [["float32", "DHWCN"], ["float32", "FRACTAL_Z_3D"]], [["float32", "FRACTAL_Z_3D"], ["float32", "DHWCN"]], [["float32", "NDC1HWC0"], ["float32", "NDHWC"]], [["float32", "NDHWC"], ["float32", "NDC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NCDHW"]], [["float32", "NCDHW"], ["float32", "NDC1HWC0"]], [["float32", "NDC1HWC0"], ["float32", "NCDHW"]], [["float32", "NCDHW"], ["float32", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "trans_data.so", "compute_cost": 10, "kernel_name": "trans_data", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "TopK", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "assist_seq", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "values", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "k", "param_type": "required", "type": "int", "value": "all"}, {"name": "sorted", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "dim", "param_type": "optional", "type": "int", "value": "all", "default_value": "-1"}, {"name": "largest", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "top_k_d.so", "compute_cost": 10, "kernel_name": "top_k_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "MatMul", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "bias", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 3, "name": "offset_w", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "transpose_x1", "param_type": "required", "type": "bool", "value": "all"}, {"name": "transpose_x2", "param_type": "required", "type": "bool", "value": "all"}, {"name": "offset_x", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "DefaultFormat"], ["int8", "DefaultFormat"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "DefaultFormat"], ["int8", "DefaultFormat"], ["float32", "FRACTAL_NZ"]], [["float32", "NHWC"], ["float32", "NHWC"], ["float32", "NHWC"], ["int8", "DefaultFormat"], ["float32", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int8", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "NHWC"], ["int32", "NHWC"], ["int32", "NHWC"], ["int8", "DefaultFormat"], ["int32", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "mat_mul.so", "compute_cost": 10, "kernel_name": "mat_mul", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Sub", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sub.so", "compute_cost": 10, "kernel_name": "sub", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "Sub", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sub.so", "compute_cost": 10, "kernel_name": "sub", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "ScatterNd", "inputs": [{"index": 0, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "shape", "param_type": "optional", "type": "listInt", "value": "all"}], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_nd_d.so", "compute_cost": 10, "kernel_name": "scatter_nd_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ScatterNdD", "inputs": [{"index": 0, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "shape", "param_type": "optional", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_nd_d.so", "compute_cost": 10, "kernel_name": "scatter_nd_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "ScatterNdAdd", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_nd_add.so", "compute_cost": 10, "kernel_name": "scatter_nd_add", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "ScatterNdAdd", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_nd_add.so", "compute_cost": 10, "kernel_name": "scatter_nd_add", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ScatterNdSub", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_nd_sub.so", "compute_cost": 10, "kernel_name": "scatter_nd_sub", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ScatterNonAliasingAdd", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_non_aliasing_add.so", "compute_cost": 10, "kernel_name": "scatter_non_aliasing_add", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ReduceMean", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_mean.so", "compute_cost": 10, "kernel_name": "reduce_mean_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
@@ -189,7 +246,7 @@
 {"op_name": "BNTrainingUpdate", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "sum", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "square_sum", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "scale", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "offset", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "batch_mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "batch_variance", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "factor", "param_type": "optional", "type": "float", "value": "all"}, {"name": "epsilon", "param_type": "optional", "type": "float", "value": "all"}, {"name": "isRef", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bn_training_update.so", "compute_cost": 10, "kernel_name": "bn_training_update", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "BNTrainingUpdateGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 2, "name": "batch_mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "batch_variance", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "diff_scale", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "diff_offset", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "epsilon", "param_type": "optional", "type": "float", "value": "all", "default_value": "0.0001"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bn_training_update_grad.so", "compute_cost": 10, "kernel_name": "bn_training_update_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "BNInfer", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "scale", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "offset", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}], "attr": [{"name": "epsilon", "param_type": "required", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bn_infer.so", "compute_cost": 10, "kernel_name": "bn_infer", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "BNInferGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "scale", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "batch_variance", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "x_backprop", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}], "attr": [{"name": "epsilon", "param_type": "optional", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bn_infer_grad.so", "compute_cost": 10, "kernel_name": "bn_infer_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "BNInferGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 1, "name": "scale", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "batch_variance", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "x_backprop", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}], "attr": [{"name": "epsilon", "param_type": "optional", "type": "float", "value": "all", "default_value": "0.0001"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bn_infer_grad.so", "compute_cost": 10, "kernel_name": "bn_infer_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "BNInference", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "scale", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "offset", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "momentum", "param_type": "optional", "type": "float", "value": "all", "default_value": "0.999"}, {"name": "epsilon", "param_type": "optional", "type": "float", "value": "all", "default_value": "0.00001"}, {"name": "use_global_stats", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "true"}, {"name": "mode", "param_type": "optional", "type": "int", "value": "all", "default_value": "1"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bninference_d.so", "compute_cost": 10, "kernel_name": "bninference_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Reciprocal", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reciprocal.so", "compute_cost": 10, "kernel_name": "reciprocal", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "ReverseV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}], "fusion_type": "ELEMWISE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reverse_v2_d.so", "compute_cost": 10, "kernel_name": "reverse_v2_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
@@ -214,6 +271,7 @@
 {"op_name": "NPUGetFloatStatus", "inputs": [{"index": 0, "name": "addr", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "data", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "n_p_u_get_float_status.so", "compute_cost": 10, "kernel_name": "n_p_u_get_float_status", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "NPUAllocFloatStatus", "inputs": [], "outputs": [{"index": 0, "name": "data", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "n_p_u_alloc_float_status.so", "compute_cost": 10, "kernel_name": "n_p_u_alloc_float_status", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "OneHot", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "on_value", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "off_value", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "depth", "param_type": "required", "type": "int", "value": "all"}, {"name": "axis", "param_type": "required", "type": "int", "value": "all", "default_value": "-1"}], "fusion_type": "OPAQUE", "dtype_format": [[["uint8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "one_hot.so", "compute_cost": 10, "kernel_name": "one_hot_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "OneHot", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "depth", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "on_value", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "off_value", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "int", "value": "all", "default_value": "-1"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "one_hot.so", "compute_cost": 10, "kernel_name": "one_hot", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Equal", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""], ["bool", ""]], [["uint8", ""], ["uint8", ""], ["bool", ""]], [["int32", ""], ["int32", ""], ["bool", ""]], [["float16", ""], ["float16", ""], ["bool", ""]], [["float32", ""], ["float32", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "equal.so", "compute_cost": 10, "kernel_name": "equal", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "Equal", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["bool", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "equal.so", "compute_cost": 10, "kernel_name": "equal", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "Less", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""], ["bool", ""]], [["uint8", ""], ["uint8", ""], ["bool", ""]], [["int32", ""], ["int32", ""], ["bool", ""]], [["float16", ""], ["float16", ""], ["bool", ""]], [["float32", ""], ["float32", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "less.so", "compute_cost": 10, "kernel_name": "less", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
@@ -222,12 +280,13 @@
 {"op_name": "LogicalNot", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "logical_not.so", "compute_cost": 10, "kernel_name": "logical_not", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "LogicalOr", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["bool", ""], ["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "logical_or.so", "compute_cost": 10, "kernel_name": "logical_or", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "ReduceMax", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]], [["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_max_d.so", "compute_cost": 10, "kernel_name": "reduce_max_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
-{"op_name": "ReduceMin", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "required", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_min_d.so", "compute_cost": 10, "kernel_name": "reduce_min_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
-{"op_name": "ReduceSum", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_sum_d.so", "compute_cost": 10, "kernel_name": "reduce_sum_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
+{"op_name": "ReduceMin", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "required", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_min_d.so", "compute_cost": 10, "kernel_name": "reduce_min_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
+{"op_name": "ReduceSum", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_sum_d.so", "compute_cost": 10, "kernel_name": "reduce_sum_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
 {"op_name": "Round", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]], [["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "round.so", "compute_cost": 10, "kernel_name": "round", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "Tanh", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "tanh.so", "compute_cost": 10, "kernel_name": "tanh", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "TanhGrad", "inputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "z", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "tanh_grad.so", "compute_cost": 10, "kernel_name": "tanh_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Softmax", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softmax.so", "compute_cost": 10, "kernel_name": "softmax_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
+{"op_name": "Softmax", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softmax.so", "compute_cost": 10, "kernel_name": "softmax_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "Softsign", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softsign.so", "compute_cost": 10, "kernel_name": "softsign", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "Softplus", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softplus.so", "compute_cost": 10, "kernel_name": "softplus", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "SoftplusGrad", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "features", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "softplus_grad.so", "compute_cost": 10, "kernel_name": "softplus_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
@@ -256,7 +315,7 @@
 {"op_name": "Minimum", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "minimum.so", "compute_cost": 10, "kernel_name": "minimum", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "Minimum", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "minimum.so", "compute_cost": 10, "kernel_name": "minimum", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "MinimumGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "grad_x", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "grad_y", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "minimum_grad.so", "compute_cost": 10, "kernel_name": "minimum_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
-{"op_name": "MaximumGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "grad_x", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "grad_y", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "maximum_grad.so", "compute_cost": 10, "kernel_name": "maximum_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
+{"op_name": "MaximumGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "grad_x", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "grad_y", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "maximum_grad.so", "compute_cost": 10, "kernel_name": "maximum_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "Concat", "inputs": [{"index": 0, "name": "input_values", "need_compile": false, "param_type": "dynamic", "shape": "all"}], "outputs": [{"index": 0, "name": "output_data", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "concat_d.so", "compute_cost": 10, "kernel_name": "concat_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "Concat", "inputs": [{"index": 0, "name": "input_values", "need_compile": false, "param_type": "dynamic", "shape": "all"}], "outputs": [{"index": 0, "name": "output_data", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "concat_d.so", "compute_cost": 10, "kernel_name": "concat_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "Slice", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "begin", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "size", "param_type": "required", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "slice_d.so", "compute_cost": 10, "kernel_name": "slice_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
@@ -265,14 +324,16 @@
 {"op_name": "ClipByNormNoDivSum", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "input3", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output_y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "clip_by_norm_no_div_sum.so", "compute_cost": 10, "kernel_name": "clip_by_norm_no_div_sum", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ClipByValue", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "clip_value_min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "clip_value_max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "dst_type", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "ELEMWISE", "dtype_format": [[["int32", ""], ["int32", ""], ["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "clip_by_value.so", "compute_cost": 10, "kernel_name": "clip_by_value", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "LayerNormBetaGammaBackprop", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "pd_gamma", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "pd_beta", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "shape_gamma", "param_type": "required", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float32", ""], ["float32", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm_beta_gamma_backprop.so", "compute_cost": 10, "kernel_name": "layer_norm_beta_gamma_backprop", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
+{"op_name": "LayerNormBetaGammaBackprop", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "pd_gamma", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "pd_beta", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "shape_gamma", "param_type": "required", "type": "listInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float32", ""], ["float32", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm_beta_gamma_backprop.so", "compute_cost": 10, "kernel_name": "layer_norm_beta_gamma_backprop", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "LayerNorm", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "gamma", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "beta", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "begin_norm_axis", "param_type": "required", "type": "int", "value": "all"}, {"name": "begin_params_axis", "param_type": "required", "type": "int", "value": "all"}, {"name": "epsilon", "param_type": "optional", "type": "float", "value": "all", "default_value": "1e-12"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm.so", "compute_cost": 10, "kernel_name": "layer_norm", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
+{"op_name": "LayerNorm", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "gamma", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "beta", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "begin_norm_axis", "param_type": "required", "type": "int", "value": "all"}, {"name": "begin_params_axis", "param_type": "required", "type": "int", "value": "all"}, {"name": "epsilon", "param_type": "optional", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm.so", "compute_cost": 10, "kernel_name": "layer_norm", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "LayerNormGrad", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "gamma", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "pd_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "pd_gamma", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "pd_beta", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm_grad.so", "compute_cost": 10, "kernel_name": "layer_norm_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "LayerNormXBackprop", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "gamma", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "pd_x", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm_x_backprop.so", "compute_cost": 10, "kernel_name": "layer_norm_x_backprop", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
+{"op_name": "LayerNormXBackprop", "inputs": [{"index": 0, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "variance", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mean", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "gamma", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "pd_x", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "layer_norm_x_backprop.so", "compute_cost": 10, "kernel_name": "layer_norm_x_backprop", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "L2Loss", "inputs": [{"index": 0, "name": "x", "param_type": "required"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "l2_loss.so", "compute_cost": 10, "kernel_name": "l2_loss", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "L2Normalize", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "epsilon", "param_type": "required", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "l2_normalize.so", "compute_cost": 10, "kernel_name": "l2_normalize", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "L2NormalizeGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dx", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "epsilon", "param_type": "required", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "l2_normalize_grad.so", "compute_cost": 10, "kernel_name": "l2_normalize_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "SquareSumV1", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "square_sum_v1.so", "compute_cost": 10, "kernel_name": "square_sum_v1", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "SquareSumV2", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "square_sum_v2.so", "compute_cost": 10, "kernel_name": "square_sum_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "SquareSumV1", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "square_sum_v1.so", "compute_cost": 10, "kernel_name": "square_sum_v1", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "SquareSumV2", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "square_sum_v2.so", "compute_cost": 10, "kernel_name": "square_sum_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ConfusionTransposeD", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "perm", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "shape", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "transpose_first", "param_type": "required", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "confusion_transpose_d.so", "compute_cost": 10, "kernel_name": "confusion_transpose_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "ConfusionSoftmaxGrad", "inputs": [{"index": 0, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "confusion_softmax_grad.so", "compute_cost": 10, "kernel_name": "confusion_softmax_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "LambUpdateWithLrV2", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "x3", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "x4", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "x5", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "greater_y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "select_e", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "lamb_update_with_lr_v2.so", "compute_cost": 10, "kernel_name": "lamb_update_with_lr_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -307,11 +368,13 @@
 {"op_name": "AvgPool", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 2, "name": "bias", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_mode", "param_type": "required", "type": "str", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}, {"name": "offset_x", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "NC1HWC0"]], [["int8", "NC1HWC0"], ["int8", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["int32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "avg_pool.so", "compute_cost": 10, "kernel_name": "avg_pool", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "AvgPoolGrad", "inputs": [{"index": 0, "name": "input_grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mean_matrix", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 2, "name": "kernel_matrix", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "out_grad", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "x_origin", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_mode", "param_type": "required", "type": "str", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "C1HWNCoC0"], ["float16", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "avg_pool_grad_d.so", "compute_cost": 10, "kernel_name": "avg_pool_grad_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "AvgPoolGradVm", "inputs": [{"index": 0, "name": "input_grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "mean_matrix", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 2, "name": "kernel_matrix", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "out_grad", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "x_origin", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_mode", "param_type": "required", "type": "str", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "C1HWNCoC0"], ["float16", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "avg_pool_grad_d.so", "compute_cost": 10, "kernel_name": "avg_pool_grad_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "AvgPool3D", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 2, "name": "multiplier", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_list", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "ceil_mode", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "count_include_pad", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "divisor_override", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all", "default_value": "NCDHW"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NDC1HWC0"], ["float16", "FRACTAL_Z_3D"], ["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "avg_pool3d_d.so", "compute_cost": 10, "kernel_name": "avg_pool3d_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "AvgPool3DGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 2, "name": "multiplier", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "output", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "origin_input_shape", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "kernel_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_list", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "ceil_mode", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "count_include_pad", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "divisor_override", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all", "default_value": "NCDHW"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NDC1HWC0"], ["float16", "FRACTAL_Z_3D"], ["float16", "NDC1HWC0"], ["float16", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "avg_pool3d_grad_d.so", "compute_cost": 10, "kernel_name": "avg_pool3d_grad_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "OnesLike", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["uint8", ""], ["uint8", ""]], [["int8", ""], ["int8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "ones_like.so", "compute_cost": 10, "kernel_name": "ones_like", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "OnesLike", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["uint8", ""], ["uint8", ""]], [["int8", ""], ["int8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "ones_like.so", "compute_cost": 10, "kernel_name": "ones_like", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "BatchToSpace", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "block_size", "param_type": "required", "type": "int", "value": "all"}, {"name": "crops", "param_type": "required", "type": "listListInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "batch_to_space_d.so", "compute_cost": 10, "kernel_name": "batch_to_space_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SpaceToBatch", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "block_size", "param_type": "required", "type": "int", "value": "all"}, {"name": "paddings", "param_type": "required", "type": "listListInt", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "space_to_batch_d.so", "compute_cost": 10, "kernel_name": "space_to_batch_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "DepthToSpace", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "block_size", "param_type": "required", "type": "int", "value": "all"}, {"name": "data_format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NHWC"], ["float16", "NHWC"]], [["float32", "NHWC"], ["float32", "NHWC"]], [["int8", "NHWC"], ["int8", "NHWC"]], [["int16", "NHWC"], ["int16", "NHWC"]], [["int32", "NHWC"], ["int32", "NHWC"]], [["int64", "NHWC"], ["int64", "NHWC"]], [["uint8", "NHWC"], ["uint8", "NHWC"]], [["uint16", "NHWC"], ["uint16", "NHWC"]], [["uint32", "NHWC"], ["uint32", "NHWC"]], [["uint64", "NHWC"], ["uint64", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "depth_to_space.so", "compute_cost": 10, "kernel_name": "depth_to_space", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "DepthToSpace", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "block_size", "param_type": "required", "type": "int", "value": "all"}, {"name": "data_format", "param_type": "optional", "type": "str", "value": "all", "default_value": "NHWC"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NHWC"], ["float16", "NHWC"]], [["float32", "NHWC"], ["float32", "NHWC"]], [["int8", "NHWC"], ["int8", "NHWC"]], [["int16", "NHWC"], ["int16", "NHWC"]], [["int32", "NHWC"], ["int32", "NHWC"]], [["int64", "NHWC"], ["int64", "NHWC"]], [["uint8", "NHWC"], ["uint8", "NHWC"]], [["uint16", "NHWC"], ["uint16", "NHWC"]], [["uint32", "NHWC"], ["uint32", "NHWC"]], [["uint64", "NHWC"], ["uint64", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "depth_to_space.so", "compute_cost": 10, "kernel_name": "depth_to_space", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SpaceToDepth", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "block_size", "param_type": "required", "type": "int", "value": "all"}, {"name": "data_format", "param_type": "optional", "type": "str", "value": "all", "default_value": "NHWC"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "FracZ"], ["float16", "NC1HWC0"]], [["float32", "NHWC"], ["float16", "FracZ"], ["float32", "NHWC"]], [["int8", "NHWC"], ["float16", "FracZ"], ["int8", "NHWC"]], [["int16", "NHWC"], ["float16", "FracZ"], ["int16", "NHWC"]], [["int32", "NHWC"], ["float16", "FracZ"], ["int32", "NHWC"]], [["int64", "NHWC"], ["float16", "FracZ"], ["int64", "NHWC"]], [["uint8", "NHWC"], ["float16", "FracZ"], ["uint8", "NHWC"]], [["uint16", "NHWC"], ["float16", "FracZ"], ["uint16", "NHWC"]], [["uint32", "NHWC"], ["float16", "FracZ"], ["uint32", "NHWC"]], [["uint64", "NHWC"], ["float16", "FracZ"], ["uint64", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "space_to_depth.so", "compute_cost": 10, "kernel_name": "space_to_depth", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ExtractImagePatches", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "ksizes", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "rates", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "padding", "param_type": "required", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NHWC"]], [["int8", "NC1HWC0"], ["int8", "NHWC"]], [["uint8", "NC1HWC0"], ["uint8", "NHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "extract_image_patches.so", "compute_cost": 10, "kernel_name": "extract_image_patches", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Sort", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all", "default_value": "-1"}, {"name": "descending", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sort.so", "compute_cost": 10, "kernel_name": "sort", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -338,7 +401,7 @@
 {"op_name": "Stack", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "dynamic", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int16", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["bool", "DefaultFormat"], ["bool", "DefaultFormat"]], [["int8", "NDHWC"], ["int8", "NDHWC"]], [["int16", "NDHWC"], ["int16", "NDHWC"]], [["int32", "NDHWC"], ["int32", "NDHWC"]], [["int64", "NDHWC"], ["int64", "NDHWC"]], [["uint8", "NDHWC"], ["uint8", "NDHWC"]], [["uint16", "NDHWC"], ["uint16", "NDHWC"]], [["uint32", "NDHWC"], ["uint32", "NDHWC"]], [["uint64", "NDHWC"], ["uint64", "NDHWC"]], [["float16", "NDHWC"], ["float16", "NDHWC"]], [["float32", "NDHWC"], ["float32", "NDHWC"]], [["bool", "NDHWC"], ["bool", "NDHWC"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "pack.so", "compute_cost": 10, "kernel_name": "pack", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Unstack", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "dynamic", "shape": "all"}], "attr": [{"name": "num", "param_type": "optional", "type": "int", "value": "all", "default_value": "None"}, {"name": "axis", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "unpack.so", "compute_cost": 10, "kernel_name": "unpack", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "ScatterUpdate", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "false"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_update.so", "compute_cost": 10, "kernel_name": "scatter_update", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "ScatterUpdate", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_update.so", "compute_cost": 10, "kernel_name": "scatter_update", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "ScatterUpdate", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "updates", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["bool", "DefaultFormat"], ["int32", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "scatter_update.so", "compute_cost": 10, "kernel_name": "scatter_update", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "PReLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "weight", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["", ""], ["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "prelu.so", "compute_cost": 10, "kernel_name": "prelu", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "PReLUGrad", "inputs": [{"index": 0, "name": "grads", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "features", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "weights", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dx", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "da", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["", ""], ["", ""], ["", ""], ["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "prelu_grad.so", "compute_cost": 10, "kernel_name": "prelu_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
 {"op_name": "BinaryCrossEntropy", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "weight", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "output", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "reduction", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["", ""], ["", ""], ["", ""], ["", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "binary_cross_entropy.so", "compute_cost": 10, "kernel_name": "binary_cross_entropy", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": true, "op_pattern": ""}
@@ -347,10 +410,10 @@
 {"op_name": "Sin", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sin.so", "compute_cost": 10, "kernel_name": "sin", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "Cos", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "cos.so", "compute_cost": 10, "kernel_name": "cos", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "Tan", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]], [["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "tan.so", "compute_cost": 10, "kernel_name": "tan", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
-{"op_name": "CumSum", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "exclusive", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "fales"}, {"name": "reverse", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "cumsum_d.so", "compute_cost": 10, "kernel_name": "cumsum_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "CumSum", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "exclusive", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "false"}, {"name": "reverse", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "cumsum_d.so", "compute_cost": 10, "kernel_name": "cumsum_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ApplyRMSProp", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "ms", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "mom", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "lr", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "ms", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "mom", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "rho", "param_type": "required", "type": "float", "value": "all"}, {"name": "momentum", "param_type": "required", "type": "float", "value": "all"}, {"name": "epsilon", "param_type": "required", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "FracZ"], ["float32", "FracZ"], ["float32", "FracZ"], ["float32", "DefaultFormat"], ["float32", "FracZ"], ["float32", "FracZ"], ["float32", "FracZ"], ["float32", "FracZ"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "DefaultFormat"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "DefaultFormat"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "apply_rms_prop.so", "compute_cost": 10, "kernel_name": "apply_rms_prop_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "CumProd", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all"}, {"name": "exclusive", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "reverse", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "cumprod_d.so", "compute_cost": 10, "kernel_name": "cumprod_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "ReduceProd", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_prod_d.so", "compute_cost": 10, "kernel_name": "reduce_prod_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": "reduce"}
+{"op_name": "CumProd", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "exclusive", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "reverse", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["uint8", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "cumprod_d.so", "compute_cost": 10, "kernel_name": "cumprod_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "ReduceProd", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["int8", ""], ["int8", ""]], [["uint8", ""], ["uint8", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_prod_d.so", "compute_cost": 10, "kernel_name": "reduce_prod_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": "reduce"}
 {"op_name": "ReciprocalGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reciprocal_grad.so", "compute_cost": 10, "kernel_name": "reciprocal_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "SqrtGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sqrt_grad.so", "compute_cost": 10, "kernel_name": "sqrt_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SqrtGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sqrt_grad.so", "compute_cost": 10, "kernel_name": "sqrt_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -366,8 +429,8 @@
 {"op_name": "BitwiseAnd", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int16", ""], ["int16", ""], ["int16", ""]], [["uint16", ""], ["uint16", ""], ["uint16", ""]], [["int32", ""], ["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bitwise_and.so", "compute_cost": 10, "kernel_name": "bitwise_and", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "BitwiseOr", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int16", ""], ["int16", ""], ["int16", ""]], [["uint16", ""], ["uint16", ""], ["uint16", ""]], [["int32", ""], ["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bitwise_or.so", "compute_cost": 10, "kernel_name": "bitwise_or", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "BitwiseXor", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int16", ""], ["int16", ""], ["int16", ""]], [["uint16", ""], ["uint16", ""], ["uint16", ""]], [["int32", ""], ["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "bitwise_xor.so", "compute_cost": 10, "kernel_name": "bitwise_xor", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
-{"op_name": "ReduceAll", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_all_d.so", "compute_cost": 10, "kernel_name": "reduce_all_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
-{"op_name": "ReduceAny", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_any_d.so", "compute_cost": 10, "kernel_name": "reduce_any_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
+{"op_name": "ReduceAll", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_all_d.so", "compute_cost": 10, "kernel_name": "reduce_all_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
+{"op_name": "ReduceAny", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "keep_dims", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["bool", ""], ["bool", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "reduce_any_d.so", "compute_cost": 10, "kernel_name": "reduce_any_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "reduce"}
 {"op_name": "SparseApplyAdagrad", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "accum", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "accum", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "lr", "param_type": "required", "type": "float", "value": "all"}, {"name": "update_slots", "param_type": "optional", "type": "bool", "value": "all"}, {"name": "use_locking", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "NHWC"], ["float32", "NHWC"], ["float32", "NHWC"], ["int32", "NHWC"], ["float32", "NHWC"], ["float32", "NHWC"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "sparse_apply_adagrad_d.so", "compute_cost": 10, "kernel_name": "sparse_apply_adagrad_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "UnsortedSegmentMin", "inputs": [{"index": 0, "name": "data", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "segment_ids", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_segments", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["int32", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float16", "FracZ"], ["int32", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["float16", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "NC1HWC0"], ["int32", "DefaultFormat"], ["float32", "NC1HWC0"]], [["float32", "FracZ"], ["int32", "DefaultFormat"], ["float32", "FracZ"]], [["float32", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["float32", "C1HWNCoC0"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "DefaultFormat"], ["int32", "NC1HWC0"]], [["int32", "FracZ"], ["int32", "DefaultFormat"], ["int32", "FracZ"]], [["int32", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["int32", "C1HWNCoC0"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "unsorted_segment_min_d.so", "compute_cost": 10, "kernel_name": "unsorted_segment_min_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "UnsortedSegmentMax", "inputs": [{"index": 0, "name": "data", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "segment_ids", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_segments", "param_type": "required", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["int32", "DefaultFormat"], ["float16", "NC1HWC0"]], [["float16", "FracZ"], ["int32", "DefaultFormat"], ["float16", "FracZ"]], [["float16", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["float16", "C1HWNCoC0"]], [["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "NC1HWC0"], ["int32", "DefaultFormat"], ["float32", "NC1HWC0"]], [["float32", "FracZ"], ["int32", "DefaultFormat"], ["float32", "FracZ"]], [["float32", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["float32", "C1HWNCoC0"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int32", "NC1HWC0"], ["int32", "DefaultFormat"], ["int32", "NC1HWC0"]], [["int32", "FracZ"], ["int32", "DefaultFormat"], ["int32", "FracZ"]], [["int32", "C1HWNCoC0"], ["int32", "DefaultFormat"], ["int32", "C1HWNCoC0"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "unsorted_segment_max_d.so", "compute_cost": 10, "kernel_name": "unsorted_segment_max_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": true, "is_dynamic_format": false, "op_pattern": ""}
@@ -386,9 +449,9 @@
 {"op_name": "Invert", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["int16", ""], ["int16", ""]], [["uint16", ""], ["uint16", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "invert.so", "compute_cost": 10, "kernel_name": "invert", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "BasicLSTMCell", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "c", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "w", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "CN"}, {"index": 4, "name": "b", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "mask", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "ct", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "ht", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "it", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 3, "name": "jt", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "ft", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 5, "name": "ot", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 6, "name": "tanhct", "need_compile": false, "param_type": "optional", "shape": "all"}], "attr": [{"name": "keep_prob", "param_type": "optional", "type": "float", "value": "all"}, {"name": "forget_bias", "param_type": "optional", "type": "float", "value": "all"}, {"name": "state_is_tuple", "param_type": "optional", "type": "bool", "value": "true"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float32", "DefaultFormat"], ["uint8", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float16", "DefaultFormat"], ["uint8", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "BasicLSTMCellCStateGradV2", "inputs": [{"index": 0, "name": "c", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dht", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "dct", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "it", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "jt", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "ft", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "ot", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 8, "name": "tanhct", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dct_1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "forget_bias", "param_type": "optional", "type": "float", "value": "all", "default_value": "1"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all", "default_value": "None"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_c_state_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_c_state_grad_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "BasicLSTMCellCStateGrad", "inputs": [{"index": 0, "name": "c", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dht", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dct", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "it", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "jt", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "ft", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "ot", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "tanhct", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dct_1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "forget_bias", "param_type": "optional", "type": "float", "value": "all"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_c_state_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_c_state_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "BasicLSTMCellCStateGrad", "inputs": [{"index": 0, "name": "c", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dht", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dct", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "it", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "jt", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "ft", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "ot", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "tanhct", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dct_1", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "forget_bias", "param_type": "optional", "type": "float", "value": "all", "default_value": "1"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all", "default_value": "None"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_c_state_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_c_state_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "BasicLSTMCellWeightGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "dw", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "CN"}, {"index": 1, "name": "db", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float32", "DefaultFormat"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_weight_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_weight_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "BasicLSTMCellInputGrad", "inputs": [{"index": 0, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "w", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 2, "name": "dropout_mask", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "dxt", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dht", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "keep_prob", "param_type": "optional", "type": "float", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["uint8", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["uint8", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_input_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_input_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "BasicLSTMCellInputGrad", "inputs": [{"index": 0, "name": "dgate", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "w", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "NC"}, {"index": 2, "name": "dropout_mask", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "dxt", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dht", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "keep_prob", "param_type": "optional", "type": "float", "value": "all", "default_value": "1.0"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["uint8", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["uint8", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "basic_lstm_cell_input_grad.so", "compute_cost": 10, "kernel_name": "basic_lstm_cell_input_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "DynamicRNN", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "w", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "CN"}, {"index": 2, "name": "b", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "seq_length", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "init_h", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 5, "name": "init_c", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 6, "name": "wci", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 7, "name": "wcf", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 8, "name": "wco", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 9, "name": "mask", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output_h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "output_c", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "i", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "j", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "f", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "o", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "tanhc", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "cell_type", "param_type": "optional", "type": "str", "value": "all", "default_value": "LSTM"}, {"name": "direction", "param_type": "optional", "type": "str", "value": "all", "default_value": "UNIDIRECTIONAL"}, {"name": "cell_depth", "param_type": "optional", "type": "int", "value": "all", "default_value": "1"}, {"name": "use_peephole", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}, {"name": "keep_prob", "param_type": "optional", "type": "float", "value": "all", "default_value": "1"}, {"name": "cell_clip", "param_type": "optional", "type": "float", "value": "all", "default_value": "-1"}, {"name": "num_proj", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "time_major", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all", "default_value": "tanh"}, {"name": "forget_bias", "param_type": "optional", "type": "float", "value": "all", "default_value": "0"}, {"name": "gate_order", "param_type": "optional", "type": "str", "value": "all", "default_value": "ijfo"}, {"name": "is_training", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float32", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["uint8", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float16", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["uint8", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["uint8", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_ZN_LSTM"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["uint8", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "dynamic_rnn.so", "compute_cost": 10, "kernel_name": "dynamic_rnn", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "DynamicGRUV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "weight_input", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "CN"}, {"index": 2, "name": "weight_hidden", "need_compile": false, "param_type": "required", "shape": "all", "reshape_type": "CN"}, {"index": 3, "name": "bias_input", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "bias_hidden", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 5, "name": "seq_length", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 6, "name": "init_h", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "output_h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "update", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 3, "name": "reset", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "new", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 5, "name": "hidden_new", "need_compile": false, "param_type": "optional", "shape": "all"}], "attr": [{"name": "direction", "param_type": "optional", "type": "str", "value": "all", "default_value": "UNIDIRECTIONAL"}, {"name": "cell_depth", "param_type": "optional", "type": "int", "value": "all", "default_value": "1"}, {"name": "keep_prob", "param_type": "optional", "type": "float", "value": "all", "default_value": "1.0"}, {"name": "cell_clip", "param_type": "optional", "type": "float", "value": "all", "default_value": "-1.0"}, {"name": "num_proj", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}, {"name": "time_major", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "activation", "param_type": "optional", "type": "str", "value": "all", "default_value": "tanh"}, {"name": "gate_order", "param_type": "optional", "type": "str", "value": "all", "default_value": "rzh"}, {"name": "reset_after", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}, {"name": "is_training", "param_type": "optional", "type": "bool", "value": "all", "default_value": "true"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float32", "DefaultFormat"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["", "DefaultFormat"], ["float32", "DefaultFormat"], ["", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["", "DefaultFormat"], ["float16", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FracZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "dynamic_gru_v2.so", "compute_cost": 10, "kernel_name": "dynamic_gru_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "GRUV2HiddenGrad", "inputs": [{"index": 0, "name": "weight_input", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "init_h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "dy", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 4, "name": "dh", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 5, "name": "update", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 6, "name": "reset", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 7, "name": "new", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 8, "name": "hidden_new", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "dh_preh", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "dgate_h", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "dnt_x", "need_compile": false, "param_type": "optional", "shape": "all"}], "attr": [{"name": "gate_order", "param_type": "optional", "type": "str", "value": "all", "default_value": "rzh"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"], ["float32", "FRACTAL_NZ"]], [["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"], ["float16", "FRACTAL_NZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gru_v2_hidden_grad.so", "compute_cost": 10, "kernel_name": "gru_v2_hidden_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -401,7 +464,7 @@
 {"op_name": "FusedMulAddNL2loss", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "x3", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "DefaultFormat"], ["float16", "NC1HWC0"], ["float16", "DefaultFormat"]], [["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "DefaultFormat"], ["float16", "C1HWNCoC0"], ["float16", "DefaultFormat"]], [["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float16", "FracZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "FracZ"], ["float16", "DefaultFormat"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "DefaultFormat"], ["float32", "NC1HWC0"], ["float32", "DefaultFormat"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "DefaultFormat"], ["float32", "C1HWNCoC0"], ["float32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "FracZ"], ["float32", "FracZ"], ["float32", "DefaultFormat"], ["float32", "FracZ"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fused_mul_addn_l2_loss.so", "compute_cost": 10, "kernel_name": "fused_mul_addn_l2_loss", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "FusedMulApplyMomentumExtern", "inputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "accum", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "lr", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "momentum", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "var_copy", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "var", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "var_copy", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "accum", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "use_nesterov", "param_type": "optional", "type": "bool", "value": "true,false", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "DefaultFormat"], ["float16", "NC1HWC0"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"], ["float16", "NC1HWC0"]], [["float32", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "DefaultFormat"], ["float16", "C1HWNCoC0"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float16", "C1HWNCoC0"]], [["float32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "FracZ"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "FracZ"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "FracZ"], ["float32", "FracZ"], ["float16", "FracZ"], ["float16", "FracZ"]], [["float32", "NC1HWC0"], ["float32", "NC1HWC0"], ["float32", "DefaultFormat"], ["float32", "NC1HWC0"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "NC1HWC0"], ["float32", "NC1HWC0"], ["float16", "NC1HWC0"], ["float32", "NC1HWC0"]], [["float32", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float32", "DefaultFormat"], ["float32", "C1HWNCoC0"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "C1HWNCoC0"], ["float32", "C1HWNCoC0"], ["float16", "C1HWNCoC0"], ["float32", "C1HWNCoC0"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "DefaultFormat"], ["float32", "DefaultFormat"]], [["float32", "FracZ"], ["float32", "FracZ"], ["float32", "DefaultFormat"], ["float32", "FracZ"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float16", "FracZ"], ["float32", "FracZ"], ["float16", "FracZ"], ["float32", "FracZ"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fused_mul_apply_momentum_extern.so", "compute_cost": 10, "kernel_name": "fused_mul_apply_momentum_extern", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "LambNextRight", "inputs": [{"index": 0, "name": "input_square", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input_mul2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "mul2_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "mul3_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "truediv1_recip", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "add2_y", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y2", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "lamb_next_right.so", "compute_cost": 10, "kernel_name": "lamb_next_right", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "SparseGatherV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gather_v2_d.so", "compute_cost": 10, "kernel_name": "gather_v2_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "SparseGatherV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "axis", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"], ["uint64", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gather_v2_d.so", "compute_cost": 10, "kernel_name": "gather_v2_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "SparseGatherV2", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "indices", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "axis", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]], [["float16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"]], [["int8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int8", "DefaultFormat"]], [["uint8", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint8", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["uint32", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint32", "DefaultFormat"]], [["int16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int16", "DefaultFormat"]], [["uint16", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint16", "DefaultFormat"]], [["int64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["int64", "DefaultFormat"]], [["uint64", "DefaultFormat"], ["int64", "DefaultFormat"], ["int32", "DefaultFormat"], ["uint64", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "gather_v2.so", "compute_cost": 10, "kernel_name": "gather_v2", "partial_flag": true, "reshape_type": "", "dynamic_shape": true, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "DataFormatDimMap", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "dst_format", "param_type": "optional", "type": "str", "value": "all"}, {"name": "src_format", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["int32", "NC1HWC0"], ["int32", "NC1HWC0"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "data_format_dim_map.so", "compute_cost": 10, "kernel_name": "data_format_dim_map", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "HistogramFixedWidth", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "range", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "nbins", "param_type": "required", "type": "int", "value": "all"}, {"name": "dtype", "param_type": "optional", "type": "str", "value": "all", "default_value": "int32"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["int32", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"]], [["int32", "DefaultFormat"], ["int32", "DefaultFormat"], ["int32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "histogram_fixed_width_d.so", "compute_cost": 10, "kernel_name": "histogram_fixed_width_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -433,7 +496,7 @@
 {"op_name": "ActsULQInputGrad", "inputs": [{"index": 0, "name": "y_grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "clamp_min_mask", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "clamp_max_mask", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "x_grad", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"], ["bool", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "acts_ulq_input_grad.so", "compute_cost": 10, "kernel_name": "acts_ulq_input_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ActULQClampMinGrad", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input_y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input_z", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["bool", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "act_ulq_clamp_min_grad.so", "compute_cost": 10, "kernel_name": "act_ulq_clamp_min_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "ActULQClampMaxGrad", "inputs": [{"index": 0, "name": "input_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input_y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input_z", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["bool", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["bool", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "act_ulq_clamp_max_grad.so", "compute_cost": 10, "kernel_name": "act_ulq_clamp_max_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "WtsARQ", "inputs": [{"index": 0, "name": "w", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "w_min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "w_max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_bits", "param_type": "optional", "type": "int", "value": "all"}, {"name": "offset_flag", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "wts_arq.so", "compute_cost": 10, "kernel_name": "wts_arq", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "WtsARQ", "inputs": [{"index": 0, "name": "w", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "w_min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "w_max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_bits", "param_type": "optional", "type": "int", "value": "all", "default_value": "8"}, {"name": "offset_flag", "param_type": "optional", "type": "bool", "value": "all", "default_value": "false"}], "fusion_type": "OPAQUE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "wts_arq.so", "compute_cost": 10, "kernel_name": "wts_arq", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "FakeQuantWithMinMaxVars", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_bits", "param_type": "optional", "type": "int", "value": "all"}, {"name": "narrow_range", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fake_quant_with_min_max_vars.so", "compute_cost": 10, "kernel_name": "fake_quant_with_min_max_vars", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "FakeQuantWithMinMaxVarsGradient", "inputs": [{"index": 0, "name": "gradients", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "backprops_wrt_x", "need_compile": true, "param_type": "required", "shape": "all"}, {"index": 1, "name": "backprops_wrt_min", "need_compile": true, "param_type": "required", "shape": "all"}, {"index": 2, "name": "backprops_wrt_max", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_bits", "param_type": "optional", "type": "int", "value": "all"}, {"name": "narrow_range", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fake_quant_with_min_max_vars_gradient.so", "compute_cost": 10, "kernel_name": "fake_quant_with_min_max_vars_gradient", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "FakeQuantWithMinMaxVarsPerChannel", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "min", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "max", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "num_bits", "param_type": "optional", "type": "int", "value": "all"}, {"name": "narrow_range", "param_type": "optional", "type": "bool", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "fake_quant_with_min_max_vars_per_channel.so", "compute_cost": 10, "kernel_name": "fake_quant_with_min_max_vars_per_channel", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
@@ -444,8 +507,8 @@
 {"op_name": "Conv3DTranspose", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "filter", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "bias", "need_compile": false, "param_type": "optional", "shape": "all"}, {"index": 3, "name": "offset_w", "need_compile": false, "param_type": "optional", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [{"name": "input_size", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "strides", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "pad_list", "param_type": "required", "type": "listInt", "value": "all"}, {"name": "dilations", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "groups", "param_type": "optional", "type": "int", "value": "all"}, {"name": "format", "param_type": "optional", "type": "str", "value": "all"}, {"name": "output_padding", "param_type": "optional", "type": "listInt", "value": "all"}, {"name": "offset_x", "param_type": "optional", "type": "int", "value": "all", "default_value": "0"}], "fusion_type": "CONVLUTION", "dtype_format": [[["float16", "NDC1HWC0"], ["float16", "FRACTAL_Z_3D"], ["float16", "DefaultFormat"], ["int8", "DefaultFormat"], ["float16", "NDC1HWC0"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "conv3d_transpose_d.so", "compute_cost": 10, "kernel_name": "conv3d_transpose_d", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "LambApplyOptimizerAssign", "inputs": [{"index": 0, "name": "grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "inputv", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "inputm", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "input3", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "mul0_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 5, "name": "mul1_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 6, "name": "mul2_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 7, "name": "mul3_x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 8, "name": "add2_y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 9, "name": "steps", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 10, "name": "do_use_weight", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 11, "name": "weight_decay_rate", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "output0", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 0, "name": "inputv", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 0, "name": "inputm", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "lamb_apply_optimizer_assign.so", "compute_cost": 10, "kernel_name": "lamb_apply_optimizer_assign", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "LambApplyWeightAssign", "inputs": [{"index": 0, "name": "input0", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "input1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "input2", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "input3", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "input_param", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "input_param", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"], ["float16", "DefaultFormat"]], [["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "lamb_apply_weight_assign.so", "compute_cost": 10, "kernel_name": "lamb_apply_weight_assign", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "NLLLoss", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "weight", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "total_weight", "need_compile": false, "param_type": "optional", "shape": "all"}], "attr": [{"name": "reduction", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "nll_loss.so", "compute_cost": 10, "kernel_name": "nll_loss", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
-{"op_name": "NLLLossGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y_grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "weight", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "total_weight", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "x_grad", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "reduction", "param_type": "optional", "type": "str", "value": "all"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "nll_loss_grad.so", "compute_cost": 10, "kernel_name": "nll_loss_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "NLLLoss", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "weight", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "total_weight", "need_compile": false, "param_type": "optional", "shape": "all"}], "attr": [{"name": "reduction", "param_type": "optional", "type": "str", "value": "all", "default_value": "mean"}, {"name": "ignore_index", "param_type": "optional", "type": "int", "value": "all", "default_value": "-100"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "nll_loss.so", "compute_cost": 10, "kernel_name": "nll_loss", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
+{"op_name": "NLLLossGrad", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "y_grad", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 2, "name": "target", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 3, "name": "weight", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 4, "name": "total_weight", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "x_grad", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [{"name": "reduction", "param_type": "optional", "type": "str", "value": "all", "default_value": "mean"}, {"name": "ignore_index", "param_type": "optional", "type": "int", "value": "all", "default_value": "-100"}], "fusion_type": "OPAQUE", "dtype_format": [[["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["int32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"], ["float32", "DefaultFormat"]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "nll_loss_grad.so", "compute_cost": 10, "kernel_name": "nll_loss_grad", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": ""}
 {"op_name": "Mish", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "mish.so", "compute_cost": 10, "kernel_name": "mish", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}
 {"op_name": "MulNoNan", "inputs": [{"index": 0, "name": "x1", "need_compile": false, "param_type": "required", "shape": "all"}, {"index": 1, "name": "x2", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": false, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["float16", ""], ["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""], ["float32", ""]], [["int32", ""], ["int32", ""], ["int32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "mul_no_nan.so", "compute_cost": 10, "kernel_name": "mul_no_nan", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "broadcast"}
 {"op_name": "SeLU", "inputs": [{"index": 0, "name": "x", "need_compile": false, "param_type": "required", "shape": "all"}], "outputs": [{"index": 0, "name": "y", "need_compile": true, "param_type": "required", "shape": "all"}], "attr": [], "fusion_type": "ELEMWISE", "dtype_format": [[["int8", ""], ["int8", ""]], [["int32", ""], ["int32", ""]], [["float16", ""], ["float16", ""]], [["float32", ""], ["float32", ""]]], "imply_type": "TBE", "async_flag": false, "binfile_name": "selu.so", "compute_cost": 10, "kernel_name": "selu", "partial_flag": true, "reshape_type": "", "dynamic_shape": false, "need_check_supported": false, "is_dynamic_format": false, "op_pattern": "formatAgnostic"}