a18949446876
/
mindspore2022
forked from huawei/mindspore2022
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
42,204 Commits 18 Branches 22 Tags 3.5 GiB
Commit Graph

15 Commits

Author SHA1 Message Date
b00518648 dbc8ba0abc add accum_loss for api 2022-03-31 11:07:39 +08:00
b00518648 87e13c5458 fix some issues of chinese api 2022-03-29 22:00:57 +08:00
juzhong3 20741d83cb change for ch api_python CrossEntropyLoss 2022-03-20 18:20:39 +08:00
lvmingfu 64e346be67 fix api name 2022-03-10 11:15:23 +08:00
i-robot 76c43e0bf8
!30484 docs_moe 
Merge pull request !30484 from wangshengnan123/code_docs_moe
 2022-03-09 03:50:28 +00:00
wangshengnan123 2872f99ea2 add chinese api 2022-03-09 09:47:17 +08:00
xumengjuan1 a758f7cb84 modify format 2022-03-08 14:34:16 +08:00
xumengjuan1 5f6b6d0970 modify apiformat 2022-03-04 15:40:30 +08:00
huangxinjing 158b5b474e Add Chinese API 2022-03-02 10:33:12 +08:00
huodagu 1d48f7387b 输入参数校准v1 2022-02-28 15:04:55 +08:00
wangshengnan12@huawei.com bb761a47be add_TransformerRecomputeConfig_API 2022-02-25 11:18:51 +08:00
Xiaoda Zhang b714451937 implementing expert_parallel+data_parallel in MoE: 
1) extending _Linear's input as 4-dimension tensor: [outer_batch, expert_dim, -1, hidden], and _Liner's BatchMatMul becomes BatchMatMul(4_dim_tensor, 3_dim_tensor);
2) configuring the _Linear's BatchMatMul sharding strategy as [[dp, ep, 1, 1], [ep, 1, mp]];
3) introducing a new parameter 'expert_parallel' in TransformerOpParallelConfig, creating a new class MoEParallelConfig to include 'data_parallel', 'model_parallel' and 'expert_parallel';
4) changing parallel config for FeedForward, TransformerEncoderLayer, TransformerDecoderLayer.
 2022-02-24 09:24:27 +08:00
zhangyi 5d8887baab modify error format for api. 2022-02-22 17:02:04 +08:00
zhangyi eddb5f9ca0 modify error format for chinese python api. 2022-02-21 19:01:22 +08:00
wangshengnan12@huawei.com 3a27ee3533 code_docs_add_transformer_api 2022-02-21 10:41:16 +08:00