Skip to content

test: add phi-4 multimodel and bielik-11b-v2.2 models for perf test #5826

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

Merged
merged 3 commits into from
Jul 21, 2025
Merged

test: add phi-4 multimodel and bielik-11b-v2.2 models for perf test #5826

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
146544e
add phi-4 multimodel and bielik-11b-v2.2 models
ruodil Jul 8, 2025
a16d8c0
add lora configs
ruodil Jul 18, 2025
bdf7377
Merge branch 'main' into user/ruodil/add_cases
LarryXFly Jul 21, 2025
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
11 changes: 11 additions & 0 deletions tests/integration/defs/perf/pytorch_model_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -186,6 +186,17 @@ def get_model_yaml_config(model_label: str,
'max_lora_rank': 64
}
}
if 'phi_4_multimodal_instruct' in model_label:
lora_config['lora_config']['lora_target_modules'] = [
"attn_qkv", "attn_dense", "mlp_h_to_4h", "mlp_4h_to_h"
]
lora_config['lora_config']['trtllm_modules_to_hf_modules'] = {
"attn_qkv": "qkv_proj",
"attn_dense": "o_proj",
"mlp_h_to_4h": "gate_up_proj",
"mlp_4h_to_h": "down_proj"
}
lora_config['lora_config']['max_lora_rank'] = 64
base_config.update(lora_config)

kv_cache_config = base_config.get('kv_cache_config', KvCacheConfig())
Expand Down
21 changes: 17 additions & 4 deletions tests/integration/defs/perf/test_perf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,6 +114,11 @@
"phi_3_mini_4k_instruct": "Phi-3/Phi-3-mini-4k-instruct",
"phi_3_mini_128k_instruct": "Phi-3/Phi-3-mini-128k-instruct",
"phi_4_mini_instruct": "Phi-4-mini-instruct",
"phi_4_multimodal_instruct": "multimodals/Phi-4-multimodal-instruct",
"phi_4_multimodal_instruct_image": "multimodals/Phi-4-multimodal-instruct",
"phi_4_multimodal_instruct_audio": "multimodals/Phi-4-multimodal-instruct",
"bielik_11b_v2.2_instruct": "Bielik-11B-v2.2-Instruct",
"bielik_11b_v2.2_instruct_fp8": "Bielik-11B-v2.2-Instruct-FP8",
}
# Model PATH of HuggingFace
HF_MODEL_PATH = {
Expand Down Expand Up @@ -145,11 +150,18 @@
"phi_4_mini_instruct_hf": "microsoft/Phi-4-mini-instruct",
}
LORA_MODEL_PATH = {
"llama_v2_13b": "llama-models-v2/chinese-llama-2-lora-13b",
"mixtral_8x7b_0.1": "chinese-mixtral-lora",
"llama_v3.1_8b_instruct_fp8": "lora/llama-3-chinese-8b-instruct-v2-lora/",
"llama_v2_13b":
"llama-models-v2/chinese-llama-2-lora-13b",
"mixtral_8x7b_0.1":
"chinese-mixtral-lora",
"llama_v3.1_8b_instruct_fp8":
"lora/llama-3-chinese-8b-instruct-v2-lora/",
"ministral_8b":
"lora/ministral/Ministral-8B-Instruct-2410-Loras-Dummy", # Dummy LoRA for Ministral
"phi_4_multimodal_instruct_image":
"multimodals/Phi-4-multimodal-instruct/vision-lora",
"phi_4_multimodal_instruct_audio":
"multimodals/Phi-4-multimodal-instruct/speech-lora",
}

TIMING_CACHE_DIR = os.environ.get("TIMING_CACHE_DIR", "")
Expand Down Expand Up @@ -1245,7 +1257,8 @@ def get_trtllm_bench_command(self, engine_dir):
#use default yaml config
if self._config.backend == "pytorch":
import yaml
config = get_model_yaml_config(self._config.to_string())
config = get_model_yaml_config(self._config.to_string(),
lora_dirs=self.lora_dirs)
print_info(f"pytorch model config: {config}")
with open('extra-llm-api-config.yml', 'w') as f:
yaml.dump(config, f, default_flow_style=False)
Expand Down
16 changes: 15 additions & 1 deletion tests/integration/test_lists/qa/trt_llm_release_perf_test.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,16 @@ trt_llm_release_perf_test:
# reduced 'reqs' to fit timeout limit
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:500,2000-reqs:8-con:1]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:500,2000-quant:fp8-reqs:8-con:1]
# Phi-4-multimodal-instruct
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct-bench-pytorch-bfloat16-input_output_len:500,2000-con:250]
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct-bench-pytorch-bfloat16-input_output_len:1000,1000-con:250]
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct-bench-pytorch-bfloat16-input_output_len:128,128]
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct-bench-pytorch-bfloat16-input_output_len:512,32]
# Bielik-11B-v2.2-Instruct
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct-bench-pytorch-bfloat16-input_output_len:128,128]
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct-bench-pytorch-bfloat16-input_output_len:512,32]
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct-bench-pytorch-bfloat16-input_output_len:1000,1000-con:250]
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct-bench-pytorch-bfloat16-input_output_len:2000,2000-con:250]
# Test list validation
- test_list_validation.py::test_list_validation

Expand All @@ -89,7 +99,9 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-float16-mp-input_output_len:128,128+512,32] #oom for l40s, l20（cuda_runtime_error）#44, mpi abort on a100 36
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-bfloat16-gwp:0.0-input_output_len:128,128+512,32] #oom for l40s, l20, mpi abort on a100 35
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-bfloat16-gwp:0.5-input_output_len:128,128+512,32] #oom for l40s, l20
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-maxnt:5000-input_output_len:5000,500-reqs:10-con:1] # timeout for l20, l40s
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:5000,500-reqs:10-con:1] # timeout for l20, l40s
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct_image-bench-pytorch-bfloat16-input_output_len:1000,1000-loras:1-con:250]
- perf/test_perf.py::test_perf[phi_4_multimodal_instruct_audio-bench-pytorch-bfloat16-input_output_len:1000,1000-loras:1-con:250]

# Llama-3.1-Nemotron-Nano-8B-v1
# cpp backend
Expand Down Expand Up @@ -158,6 +170,8 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-float16-maxbs:256-input_output_len:500,2000-quant:fp8]
- perf/test_perf.py::test_perf[phi_3_mini_4k_instruct-bench-float16-maxbs:128-input_output_len:1000,1000-quant:fp8]
- perf/test_perf.py::test_perf[phi_3_mini_4k_instruct-bench-float16-maxbs:64-input_output_len:500,2000-quant:fp8]
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct_fp8-bench-pytorch-float8-input_output_len:1000,1000-con:250]
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct_fp8-bench-pytorch-float8-input_output_len:2000,2000-con:250]

- condition:
terms:
Expand Down