From cea8e1f5e491f6c5b1f49ba5e02fdc10c734cc5e Mon Sep 17 00:00:00 2001 From: Taylor Yeonbok Lee Date: Mon, 3 Jun 2024 03:34:13 -0700 Subject: [PATCH] [GPU] Fixed unexpected kv cache optimization turned off (#24805) ### Details: - kv cache opt was disabled due to the wrongly detected beam table size. Fixed this issue. ### Tickets: - *ticket-id* --- src/plugins/intel_gpu/src/graph/primitive_inst.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp index a9d04d3fc76..2aca7c6a4dc 100644 --- a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp +++ b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp @@ -1034,8 +1034,8 @@ void primitive_inst::do_runtime_in_place_kv_cache() { const auto& sequence_axis = desc->concat_axis; const auto& gather_axis = desc->gather_axis; - const auto& prev_batch_size = past_layout.get_shape()[gather_axis]; - const auto& beam_size = present_layout.get_shape()[gather_axis]; + const auto& prev_batch_size = static_cast(past_layout.get_shape()[gather_axis]); + const auto& beam_size = static_cast(present_layout.get_shape()[gather_axis]); if (prev_batch_size != beam_size) { // If the previous batch size is not same as beam size, need explicit concat _impl_params->_can_be_optimized = false;