Skip to content

[https://nvbugs/5636946][fix] Update test model #8993

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 2 commits into from
Nov 7, 2025
Merged

[https://nvbugs/5636946][fix] Update test model #8993

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4e8080d
update sm120 deepseek case
crazydemo Nov 6, 2025
8acc07f
unwaive H20-3e cases
crazydemo Nov 7, 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
133 changes: 133 additions & 0 deletions tests/integration/defs/accuracy/test_llm_api_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2057,6 +2057,139 @@ def test_nvfp4_multi_gpus(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
# task.evaluate(llm,
# extra_evaluator_kwargs=dict(apply_chat_template=True))

@skip_pre_blackwell
@pytest.mark.parametrize(
"tp_size,pp_size,ep_size,mtp_nextn,fp8kv,attention_dp,cuda_graph,overlap_scheduler,max_batch_size,moe_backend",
[
# Use a larger batch_size to speed up the tests
pytest.param(8,
1,
4,
3,
False,
False,
True,
True,
32,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(8)),
pytest.param(8,
1,
4,
3,
False,
False,
True,
True,
32,
"TRTLLM",
marks=pytest.mark.skip_less_mpi_world_size(8)),
pytest.param(8,
1,
8,
0,
True,
True,
True,
True,
32,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(8)),
pytest.param(8,
1,
1,
0,
True,
True,
True,
True,
32,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(8)),
pytest.param(4,
1,
1,
0,
True,
True,
True,
True,
16,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(4)),
pytest.param(8,
1,
8,
1,
True,
True,
True,
True,
32,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(8)),
pytest.param(8,
1,
8,
1,
True,
True,
True,
True,
8,
"CUTLASS",
marks=pytest.mark.skip_less_mpi_world_size(8)),
],
ids=[
"latency", "latency_trtllmgen", "throughput", "throughput_tp8",
"throughput_tp4", "throughput_mtp", "throughput_bs8_mtp"
])
def test_nvfp4_multi_gpus_sm120(self, tp_size, pp_size, ep_size, mtp_nextn,
fp8kv, attention_dp, cuda_graph,
overlap_scheduler, max_batch_size,
moe_backend):
if moe_backend == "TRTLLM" and (get_sm_version() == 120
or get_sm_version() == 121):
pytest.skip(
"MOE TRTLLM backend does not support SM version 120 or 121")

kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.70)
pytorch_config = dict(
disable_overlap_scheduler=not overlap_scheduler,
cuda_graph_config=CudaGraphConfig() if cuda_graph else None,
moe_config=MoeConfig(backend=moe_backend))

if fp8kv:
kv_cache_config.dtype = "fp8"

mtp_config = None
if mtp_nextn > 0:
mtp_config = MTPDecodingConfig(num_nextn_predict_layers=mtp_nextn)
with LLM(f"{llm_models_root()}/DeepSeek-R1/DeepSeek-R1-0528-FP4-v2",
max_batch_size=max_batch_size,
tensor_parallel_size=tp_size,
pipeline_parallel_size=pp_size,
moe_expert_parallel_size=ep_size,
kv_cache_config=kv_cache_config,
**pytorch_config,
enable_attention_dp=attention_dp,
speculative_config=mtp_config) as llm:

assert llm.args.moe_config.backend == moe_backend
assert llm.args.quant_config.quant_algo == QuantAlgo.NVFP4

task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)
# This covers the case with relatively large seqlen in the generation phase.
task = CnnDailymail(self.MODEL_NAME)
task.evaluate(llm)
# Commented out because GPQA takes too long to run
# task = GPQADiamond(self.MODEL_NAME)
# task.evaluate(llm,
# extra_evaluator_kwargs=dict(apply_chat_template=True))

@skip_pre_blackwell
@pytest.mark.parametrize(
"tp_size,pp_size,ep_size,mtp_nextn,fp8kv,attention_dp,cuda_graph,overlap_scheduler,max_batch_size,moe_backend",
Expand Down
2 changes: 1 addition & 1 deletion tests/integration/test_lists/qa/llm_function_rtx6k.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtl
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm]
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_eagle3]
accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8]
accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus_sm120[throughput_tp8]
test_e2e.py::test_ptp_quickstart_advanced_mixed_precision
test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B]
test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]
Expand Down
3 changes: 0 additions & 3 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -339,9 +339,6 @@ triton_server/test_triton_llm.py::test_llmapi_backend[1-0-disableDecoupleMode-te
triton_server/test_triton_llm.py::test_llmapi_backend[1-0-enableDecoupleMode-tensorrt_llm] SKIP (https://nvbugs/5461874)
cpp/test_e2e.py::test_benchmarks[gpt-80] SKIP (https://nvbugs/5601670)
disaggregated/test_disaggregated.py::test_disaggregated_benchmark_on_diff_backends[llama-v3-8b-hf] SKIP (https://nvbugs/5587574)
full:H20-3e/accuracy/test_llm_api_pytorch.py::TestNemotronUltra::test_auto_dtype[tp8ep4-cuda_graph=True] SKIP (slow I/O)
full:H20-3e/accuracy/test_llm_api_pytorch.py::TestKimiK2::test_fp8_blockscale[latency] SKIP (slow I/O)
full:H20-3e/test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[DeepSeek-V3-671B-FP8-DeepSeek-V3-0324-8] SKIP (slow I/O)
disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[False-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct] SKIP (https://nvbugs/5608743)
accuracy/test_cli_flow.py::TestMinitron4BBase::test_fp8 SKIP (https://nvbugs/5606233)
examples/test_gpt.py::test_llm_minitron_fp8_with_pseudo_loras[4b] SKIP (https://nvbugs/5606233)
Expand Down