Skip to content

[TRTLLM-4932] Add LLM API accuracy tests for Llama-4-Maverick-17B-128E-Instruct - FP8 variant #4389

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

Closed
moraxu wants to merge 5 commits into from
Closed

[TRTLLM-4932] Add LLM API accuracy tests for Llama-4-Maverick-17B-128E-Instruct - FP8 variant #4389

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9e8aa8e
Add accuracy CLI TestLlama4MaverickInstruct test
moraxu May 16, 2025
f07189f
Update llama config
moraxu May 18, 2025
cf4f93f
Address review comments
moraxu May 21, 2025
9121cd9
Remove changes for trt-related configs
moraxu May 21, 2025
8c67f1e
Update the test
moraxu Jul 25, 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
3 changes: 3 additions & 0 deletions tests/integration/defs/accuracy/references/mmlu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,9 @@ meta-llama/Llama-3.3-70B-Instruct:
accuracy: 81.02
meta-llama/Llama-4-Maverick-17B-128E-Instruct:
- accuracy: 86.40
- quant_algo: FP8
kv_cache_quant_algo: FP8
accuracy: 0.0
meta-llama/Llama-4-Scout-17B-16E-Instruct:
- accuracy: 80.00
- quant_algo: NVFP4
Expand Down
19 changes: 19 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 @@ -408,6 +408,25 @@ def test_auto_dtype(self, cuda_graph, tp_size, pp_size, ep_size):
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)

@skip_pre_hopper
@pytest.mark.skip_less_device(8)
@parametrize_with_ids("cuda_graph", [False, True])
@pytest.mark.parametrize("tp_size,pp_size,ep_size", [(8, 1, 1), (8, 1, 4),
(8, 1, 8)],
ids=["tp8", "tp8ep4", "tp8ep8"])
def test_fp8_prequantized(self, cuda_graph, tp_size, pp_size, ep_size):
Comment thread
moraxu marked this conversation as resolved.
Outdated
model_path = f"{llm_models_root()}/llama4-models/nvidia/Llama-4-Maverick-17B-128E-Instruct-FP8"
with LLM(model_path,
tensor_parallel_size=tp_size,
pipeline_parallel_size=pp_size,
moe_expert_parallel_size=ep_size,
cuda_graph_config=CudaGraphConfig()
if cuda_graph else None) as llm:
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)
Comment thread
moraxu marked this conversation as resolved.


class TestLlama4ScoutInstruct(LlmapiAccuracyTestHarness):
MODEL_NAME = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
Expand Down
3 changes: 3 additions & 0 deletions tests/integration/test_lists/qa/llm_sanity_test.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,9 @@ accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4
accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_auto_dtype[tp8-cuda_graph=False]
accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_auto_dtype[tp8ep4-cuda_graph=True]
accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_auto_dtype[tp8ep8-cuda_graph=True]
accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_fp8_prequantized[tp8-cuda_graph=False]
accuracy/test_cli_flow.py::TestLlama4MaverickInstruct::test_auto_dtype[tp8-cuda_graph=False]
accuracy/test_cli_flow.py::TestLlama4MaverickInstruct::test_fp8_prequantized[tp8-cuda_graph=False]
accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8-cuda_graph=False]
accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8ep4-cuda_graph=True]
accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8ep8-cuda_graph=True]
Expand Down