kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-13 22:18:36 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[https://nvbugs/5698434][test] Add Qwen3-4B-Eagle3 One-model perf test (#10041)

Browse Source 
Signed-off-by: yufeiwu-nv <230315618+yufeiwu-nv@users.noreply.github.com>
This commit is contained in:
yufeiwu-nv 2025-12-17 13:37:25 +08:00 committed by GitHub
parent 47404196fa
commit 5d71f662c3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 19 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
tests/integration/defs/perf/pytorch_model_config.py
View File

@ -215,6 +215,23 @@ def get_model_yaml_config(model_label: str,
}
}
},
{
'patterns': [
'qwen3_4b-bench-pytorch-streaming-bfloat16-maxbs:4-kv_frac:0.6-input_output_len:500,100-reqs:200-con:4',
],
'config': {
'speculative_config': {
'decoding_type': 'Eagle',
'eagle3_one_model': True,
'speculative_model_dir': 'Qwen3-4B_eagle3',
'max_draft_len': 3,
},
'kv_cache_config': {
'enable_block_reuse': False,
},
'enable_chunked_prefill': False,
}
},
# Llama-v3.3 models with fp8 quantization
{
'patterns': [

1
tests/integration/defs/perf/test_perf.py
View File

@ -110,6 +110,7 @@ MODEL_PATH_DICT = {
"deepseek_v3_lite_nvfp4": "DeepSeek-V3-Lite/nvfp4_moe_only",
"qwen2_7b_instruct": "Qwen2-7B-Instruct",
"qwen_14b_chat": "Qwen-14B-Chat",
"qwen3_4b_eagle3": "Qwen3/Qwen3-4B",
"qwen3_235b_a22b_fp8": "Qwen3/saved_models_Qwen3-235B-A22B_fp8_hf",
"qwen3_235b_a22b_fp4": "Qwen3/saved_models_Qwen3-235B-A22B_nvfp4_hf",
"starcoder2_3b": "starcoder2-3b",

1
tests/integration/test_lists/qa/llm_perf_sanity.yml
View File

@ -27,6 +27,7 @@ llm_perf_sanity:
- perf/test_perf.py::test_perf[ministral_8b-bench-pytorch-bfloat16-input_output_len:500,2000-reqs:500-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-pytorch-bfloat16-input_output_len:500,2000]
- perf/test_perf.py::test_perf[nemotron_nano_9b_v2-bench-pytorch-bfloat16-input_output_len:512,512]
- perf/test_perf.py::test_perf[qwen3_4b_eagle3-bench-pytorch-streaming-bfloat16-maxbs:4-kv_frac:0.6-input_output_len:500,100-reqs:200-con:4]
# FP8 specific tests