add excludes for batch_size

intel · Feb 5, 2024 · d49a094 · d49a094
1 parent df5f7b6
commit d49a094
Show file tree

Hide file tree

Showing 2 changed files with 40 additions and 0 deletions.
diff --git a/python/llm/test/benchmark/stable-version-arc-perf-test-fp8.yaml b/python/llm/test/benchmark/stable-version-arc-perf-test-fp8.yaml
@@ -18,3 +18,25 @@ in_out_pairs:
 test_api:
   - "transformer_int4_gpu"  # on Intel GPU
 cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)
+exclude:
+  - 'meta-llama/Llama-2-7b-chat-hf:2048:4'
+  - 'meta-llama/Llama-2-7b-chat-hf:512:8'
+  - 'meta-llama/Llama-2-7b-chat-hf:1024:8'
+  - 'meta-llama/Llama-2-7b-chat-hf:2048:8'
+  - 'THUDM/chatglm2-6b:2048:8'
+  - 'THUDM/chatglm3-6b:2048:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:2'
+  - 'baichuan-inc/Baichuan2-7B-Chat:1024:4'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:4'
+  - 'baichuan-inc/Baichuan2-7B-Chat:512:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:1024:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:8'
+  - 'Qwen/Qwen-7B-Chat:2048:1'
+  - 'Qwen/Qwen-7B-Chat:1024:2'
+  - 'Qwen/Qwen-7B-Chat:2048:2'
+  - 'Qwen/Qwen-7B-Chat:512:4'
+  - 'Qwen/Qwen-7B-Chat:1024:4'
+  - 'Qwen/Qwen-7B-Chat:2048:4'
+  - 'Qwen/Qwen-7B-Chat:512:8'
+  - 'Qwen/Qwen-7B-Chat:1024:8'
+  - 'Qwen/Qwen-7B-Chat:2048:8'
diff --git a/python/llm/test/benchmark/stable-version-arc-perf-test-sym_int4.yaml b/python/llm/test/benchmark/stable-version-arc-perf-test-sym_int4.yaml
@@ -18,3 +18,21 @@ in_out_pairs:
 test_api:
   - "transformer_int4_gpu"  # on Intel GPU
 cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)
+exclude:
+  - 'meta-llama/Llama-2-7b-chat-hf:2048:4'
+  - 'meta-llama/Llama-2-7b-chat-hf:1024:8'
+  - 'meta-llama/Llama-2-7b-chat-hf:2048:8'
+  - 'THUDM/chatglm2-6b:2048:8'
+  - 'THUDM/chatglm3-6b:2048:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:2'
+  - 'baichuan-inc/Baichuan2-7B-Chat:1024:4'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:4'
+  - 'baichuan-inc/Baichuan2-7B-Chat:512:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:1024:8'
+  - 'baichuan-inc/Baichuan2-7B-Chat:2048:8'
+  - 'Qwen/Qwen-7B-Chat:2048:2'
+  - 'Qwen/Qwen-7B-Chat:1024:4'
+  - 'Qwen/Qwen-7B-Chat:2048:4'
+  - 'Qwen/Qwen-7B-Chat:512:8'
+  - 'Qwen/Qwen-7B-Chat:1024:8'
+  - 'Qwen/Qwen-7B-Chat:2048:8'