Skip to content

[https://nvbugs/6327147][fix] Lower kv_cache_config.free_gpu_memory_fraction from 0.7 to 0.5 in the… #15468

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
tensorrt-cicd wants to merge 1 commit into
base: main
Choose a base branch
from
+7 −3
Open

[https://nvbugs/6327147][fix] Lower kv_cache_config.free_gpu_memory_fraction from 0.7 to 0.5 in the… #15468

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion tests/integration/defs/accuracy/test_llm_api_autodeploy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,6 +412,12 @@ def get_default_kwargs(self,
enable_chunked_prefill=False,
attn_backend="flashinfer"):
yaml_paths, _ = _get_registry_yaml_extra(self.MODEL_NAME)
# Without chunked prefill, max_num_tokens=8192 makes the synthetic
# forward profile in resize_kv_cache underestimate runtime peaks
# (CUDA-graph capture + fp32 lm_head logits cast), causing OOM on
# 48GB cards. Lower the KV-cache fraction in that case to leave
# headroom for those post-resize allocations.
free_gpu_memory_fraction = 0.7 if enable_chunked_prefill else 0.5
config = {
"yaml_extra": yaml_paths,
"skip_tokenizer_init": False,
Expand All @@ -420,7 +426,7 @@ def get_default_kwargs(self,
# SSMs do not support cache reuse.
"kv_cache_config": {
"enable_block_reuse": False,
"free_gpu_memory_fraction": 0.7
"free_gpu_memory_fraction": free_gpu_memory_fraction,
},
# Keep max_batch_size as in the PyTorch test to avoid OOM
"max_batch_size": 128,
Expand Down
2 changes: 0 additions & 2 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -244,8 +244,6 @@ full:H20/llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_mu
full:H20/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760)
full:L40S/accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[triton-False-1] SKIP (https://nvbugs/6322045)
full:L40S/accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-1] SKIP (https://nvbugs/6322045)
full:L40S/accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-flashinfer_ssm-False] SKIP (https://nvbugs/6327147)
full:L40S/accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-False] SKIP (https://nvbugs/6327147)
full:L40S/accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=True] SKIP (https://nvbugs/6276841)
full:L40S/accuracy/test_llm_api_pytorch_multimodal.py::TestExaone4_5_33B::test_auto_dtype[forced_chunked_prefill] SKIP (https://nvbugs/6327149)
full:L40S/accuracy/test_llm_api_pytorch_multimodal.py::TestExaone4_5_33B::test_auto_dtype[full_budget] SKIP (https://nvbugs/6327149)
Expand Down
Loading