[GPU] Enable bf16 state conversions

openvinotoolkit · Nov 4, 2024 · 1a0866c · 1a0866c
1 parent 4d8ff86
commit 1a0866c
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 11 deletions.
diff --git a/src/plugins/intel_gpu/src/plugin/common_utils.cpp b/src/plugins/intel_gpu/src/plugin/common_utils.cpp
@@ -83,9 +83,11 @@ void convert_and_copy(const void* src_ptr, ov::element::Type src_et, void* dst_p
 
     // For state conversions
     CASE(ov::element::f32, ov::element::f32, float, float);
-    CASE(ov::element::f16, ov::element::f16, ov::float16, ov::float16);
     CASE(ov::element::f32, ov::element::f16, float, ov::float16);
     CASE(ov::element::f16, ov::element::f32, ov::float16, float);
+    CASE(ov::element::f16, ov::element::f16, ov::float16, ov::float16);
+    CASE(ov::element::bf16, ov::element::f32, ov::bfloat16, float);
+    CASE(ov::element::bf16, ov::element::f16, ov::bfloat16, ov::float16);
 
     OPENVINO_THROW("[GPU] Unsupported element types combination for copy: ", src_et, " -> ", dst_et);
 }

diff --git a/src/plugins/intel_gpu/src/runtime/execution_config.cpp b/src/plugins/intel_gpu/src/runtime/execution_config.cpp
@@ -212,16 +212,6 @@ void ExecutionConfig::apply_debug_options(const cldnn::device_info& info) {
             set_property(ov::hint::dynamic_quantization_group_size(debug_config->dynamic_quantize_group_size));
     }
 
-    int KVCacheCompression = 0;
-    if (const auto env_var = std::getenv("KVCacheCompression")) {
-        std::istringstream ss(env_var);
-        ss >> KVCacheCompression;
-    }
-
-    if (KVCacheCompression == 1) {
-        set_property(ov::hint::kv_cache_precision(ov::element::i8));
-    }
-
     GPU_DEBUG_IF(debug_config->use_kv_cache_compression != -1) {
         GPU_DEBUG_IF(debug_config->use_kv_cache_compression == 1) {
             set_property(ov::hint::kv_cache_precision(ov::element::i8));