Skip to content

test: change default max_batch_size to 512 in test config and dump config.json to log #3656

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 4 commits into from
Apr 22, 2025
Merged

test: change default max_batch_size to 512 in test config and dump config.json to log #3656

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b5b4955
waive failed case in perf test, change default max_batch_size to 512 …
ruodil Apr 17, 2025
add2bfe
Merge branch 'release/0.19' into user/ruodil/perf_release0.19
LarryXFly Apr 21, 2025
5b75ebf
Merge branch 'release/0.19' into user/ruodil/perf_release0.19
ruodil Apr 22, 2025
b9d178f
Merge branch 'release/0.19' into user/ruodil/perf_release0.19
LarryXFly Apr 22, 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
6 changes: 3 additions & 3 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 @@ -283,8 +283,8 @@ def __init__(
backend: str = "",
mode: str = "plugin",
data_type: str = "float16",
max_batch_size: int = 2048,
max_num_tokens: int = 8192,
max_batch_size: int = 512,
max_num_tokens: int = 2048,
gpu_weights_percent: float = -1,
batch_sizes: List[int] = [0],
input_lens: List[int] = [8],
Expand Down Expand Up @@ -601,7 +601,7 @@ def validate(self):
if self.model_name in MODEL_PATH_DICT.keys():
VALID_QUANTS = [
"", "nvfp4", "fp8", "int8_sq", "int4_awq", "w4a8_awq",
"int8_wo", "int4_wo", "full_prec"
"w4a16_awq", "int8_wo", "int4_wo", "full_prec"
]
else:
VALID_QUANTS = [
Expand Down
14 changes: 14 additions & 0 deletions tests/integration/defs/perf/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -265,6 +265,20 @@ def run_cmd(self, cmd_idx: int, venv) -> str:
benchmark_cmd = mpi_cmd + command
output += subprocess.check_output(benchmark_cmd,
env=envs).decode()
# write config.json to output log
match = re.search(r'--engine_dir=([^\s]+)', current_cmd_str)
if match:
engine_dir = match.group(1)
print_info(
f'writing config.json in {engine_dir} to output log')
with open(os.path.join(engine_dir, "config.json"),
"r") as f:
config_content = f.read()
output += "\n" + "=" * 50 + "\n"
output += "ENGINE CONFIG:\n"
output += "=" * 50 + "\n"
output += config_content
output += "\n" + "=" * 50 + "\n"
return output

def get_cmd_str(self, cmd_idx) -> List[str]:
Expand Down
11 changes: 5 additions & 6 deletions tests/integration/test_lists/qa/trt_llm_release_perf_cluster_test.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,16 +40,15 @@ trt_llm_release_perf_cluster_test:

# GB chip specific tests with high memory
- condition:
ranges:
system_gpu_count:
gte: 8
gpu_memory:
gt: 100000
wildcards:
chip: 'gb*'
gpu:
- '*b100*'
linux_distribution_name: '*'
tests:
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:512-input_output_len:128,128-quant:fp8-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-quant:fp8-reqs:10-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:384-maxnt:1536-input_output_len:1000,2000-quant:fp8-reqs:49152-con:3072-ep:8-tp:8-gpus:8] #max throughput test
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-float16-maxbs:512-input_output_len:128,128-quant:nvfp4-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-streaming-float16-maxbs:512-input_output_len:1000,1000-quant:nvfp4-con:4096-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-quant:nvfp4-reqs:10-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-float16-maxbs:384-maxnt:1536-input_output_len:1000,2000-quant:nvfp4-reqs:49152-con:3072-ep:8-tp:8-gpus:8] #max throughput test
31 changes: 6 additions & 25 deletions 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 @@ -74,7 +74,6 @@ trt_llm_release_perf_test:
tests:
- perf/test_perf.py::test_perf[llama_70b_sq_per_tensor-cppmanager-exe-plugin_ifb-float16-input_output_len:128,128+512,32-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b-cppmanager-exe-plugin_ifb-float16-mp-input_output_len:128,128+512,32-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-float16-maxbs:256-input_output_len:128,128+512,32-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128+512,32-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-streaming-float16-input_output_len:128,128-gpus:2]

Expand Down Expand Up @@ -123,8 +122,8 @@ trt_llm_release_perf_test:
tests:
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:32-input_output_len:128,128-quant:fp8-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-streaming-float16-maxbs:32-input_output_len:128,128-quant:fp8-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-reqs:10-quant:fp8-con:1-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:128-maxnt:1127-input_output_len:1000,2000-reqs:5120-quant:fp8-con:1024-ep:8-tp:8-gpus:8] #max throughput test
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-quant:fp8-reqs:10-con:1-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:128-maxnt:1127-input_output_len:1000,2000-quant:fp8-reqs:5120-con:1024-ep:8-tp:8-gpus:8] #max throughput test

# FP8 specific tests
- condition:
Expand All @@ -134,7 +133,7 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3.1_8b-cppmanager-exe-plugin_ifb-float16-maxbs:256-input_output_len:128,128-beams:4-quant:fp8]

- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-input_output_len:128,128-quant:fp8]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-input_output_len:128,128-quant:int4_awq]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-input_output_len:128,128-quant:w4a16_awq]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-input_output_len:128,128-quant:w4a8_awq]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-maxbs:256-input_output_len:128,128-quant:fp8]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-maxbs:256-input_output_len:512,32-quant:fp8]
Expand Down Expand Up @@ -176,7 +175,6 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:512,32-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_11b-bench-float16-input_output_len:512,200-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-float16-input_output_len:128,128-quant:fp8-tp:2]

- condition:
terms:
Expand All @@ -203,28 +201,11 @@ trt_llm_release_perf_test:
# GB chip specific tests
- condition:
wildcards:
chip: 'gb*'
gpu:
- '*b100*'
- '*b40*'
linux_distribution_name: '*'
tests:
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-float16-input_output_len:128,128-quant:nvfp4]
- perf/test_perf.py::test_perf[deepseek_v3_lite_nvfp4-bench-pytorch-float16-input_output_len:128,128-quant:nvfp4]
- perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float16-input_output_len:128,128-quant:fp8]

# GB200 chip specific tests
- condition:
ranges:
system_gpu_count:
gte: 8
gpu_memory:
gt: 100000
wildcards:
chip: 'gb*'
linux_distribution_name: '*'
tests:
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:512-input_output_len:128,128-quant:fp8-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-quant:fp8-reqs:10-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1-bench-pytorch-float16-maxbs:384-maxnt:1536-input_output_len:1000,2000-quant:fp8-reqs:49152-con:3072-ep:8-tp:8-gpus:8] #max throughput test
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-float16-maxbs:512-input_output_len:128,128-quant:nvfp4-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1_nvfp4-bench-pytorch-streaming-float16-maxbs:512-input_output_len:1000,1000-quant:nvfp4-con:4096-ep:8-tp:8-gpus:8]
- perf/test_perf.py::test_perf[deepseek_r1-nvfp4-bench-pytorch-float16-maxbs:1-input_output_len:1000,2000-quant:nvfp4-reqs:10-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1-nvfp4-bench-pytorch-float16-maxbs:384-maxnt:1536-input_output_len:1000,2000-quant:nvfp4-reqs:49152-con:3072-ep:8-tp:8-gpus:8] #max throughput test
17 changes: 17 additions & 0 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -453,5 +453,22 @@ examples/test_medusa.py::test_llama_medusa_1gpu[llama-v2-7b-hf] SKIP (https://nv
examples/test_medusa.py::test_llama_medusa_1gpu[llama-3.2-1b] SKIP (https://nvbugs/5219534)
examples/test_medusa.py::test_llama_medusa_1gpu[llama-3.1-8b] SKIP (https://nvbugs/5219535)
examples/test_phi.py::test_llm_phi_quantization_1gpu[Phi-4-mini-instruct-fp8-bfloat16] SKIP (https://nvbugspro.nvidia.com/bug/5226339)
examples/test_multimodal.py::test_llm_multimodal_general[neva-22b-pp:1-tp:1-bfloat16-bs:1-cpp_e2e:True-nb:1] SKIP (https://nvbugs/5227342)
disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8] SKIP (https://nvbugs/5201168)
perf/test_perf.py::test_perf[gpt_350m-cppmanager-plugin-float16-bs:32-input_output_len:60,20] SKIP (https://nvbugs/5228840)
perf/test_perf.py::test_perf[gpt_350m-cppmanager-static_batching-plugin-float16-bs:32-input_output_len:60,20] SKIP (https://nvbugs/5228840)
perf/test_perf.py::test_perf[t5-bench-float16-input_output_len:128,20] SKIP # https://nvbugspro.nvidia.com/bug/5207477
perf/test_perf.py::test_perf[flan_t5_base-bench-float16-input_output_len:128,20] SKIP
perf/test_perf.py::test_perf[flan_t5_large-bench-float16-input_output_len:128,20] SKIP
perf/test_perf.py::test_perf[flan_t5_large-bench-float16-maxbs:1-input_output_len:128,20-gpus:2] SKIP
perf/test_perf.py::test_perf[whisper_large_v3-bench-float16-input_output_len:128,20] SKIP
perf/test_perf.py::test_perf[mamba_370m-bench-float16-input_output_len:128,128] SKIP
perf/test_perf.py::test_perf[mamba_370m-bench-float16-input_output_len:512,32] SKIP
perf/test_perf.py::test_perf[mamba_2.8b-bench-float16-input_output_len:128,128] SKIP
perf/test_perf.py::test_perf[mamba_2.8b-bench-float16-input_output_len:512,32] SKIP
perf/test_perf.py::test_perf[t5-bench-float16-input_output_len:128,20-gpus:2] SKIP
perf/test_perf.py::test_perf[t5-bench-float16-maxbs:1-input_output_len:128,20-gpus:2] SKIP
perf/test_perf.py::test_perf[gpt_20b-bench-float16-maxbs:8-input_output_len:128,128-reqs:80-gpus:8] SKIP
perf/test_perf.py::test_perf[gpt_20b-bench-float16-maxbs:8-input_output_len:512,32-reqs:80-gpus:8] SKIP
full:B40/perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float16-input_output_len:128,128-quant:fp8] SKIP #https://nvbugspro.nvidia.com/bug/5150255
full:B200/perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float16-input_output_len:128,128-quant:fp8] SKIP #https://nvbugspro.nvidia.com/bug/5150255