Skip to content

Commit cc33f66

Browse files
liji-nvliji-nv
committed
[None][feat] GPT_OSS torch compile support
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
1 parent 7e88212 commit cc33f66

File tree

13 files changed

+166
-85
lines changed

13 files changed

+166
-85
lines changed

‎tensorrt_llm/_torch/custom_ops/cpp_custom_ops.py‎

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -842,3 +842,24 @@ def _(
842842
# This is a fake implementation for shape inference
843843
# The actual operation modifies fused_q and q_pe in-place
844844
return None
845+
846+
@torch.library.register_fake("trtllm::mxfp8_quantize")
847+
def _(input: torch.Tensor,
848+
isSfSwizzledLayout: bool = True,
849+
alignment: int = 32):
850+
output_shape = list(input.shape)
851+
padded_k = fp4_utils.pad_up(output_shape[-1], alignment)
852+
output_shape[-1] = padded_k
853+
output = input.new_empty(output_shape, dtype=torch.float8_e4m3fn)
854+
855+
sf_vec_size = 32
856+
if isSfSwizzledLayout:
857+
sf_size = fp4_utils.pad_up(output_shape[0], 128) * fp4_utils.pad_up(
858+
padded_k // sf_vec_size, 4)
859+
else:
860+
m = 1
861+
for i in range(len(output_shape) - 1):
862+
m *= output_shape[i]
863+
sf_size = m * (padded_k // sf_vec_size)
864+
sf = input.new_empty((sf_size, ), dtype=torch.uint8)
865+
return output, sf

‎tensorrt_llm/_torch/models/modeling_gpt_oss.py‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -187,7 +187,8 @@ def __init__(
187187
'bias': True,
188188
'swiglu_alpha': self.swiglu_alpha,
189189
'swiglu_beta': self.swiglu_beta,
190-
'swiglu_limit': self.swiglu_limit
190+
'swiglu_limit': self.swiglu_limit,
191+
'layer_idx': self.layer_idx,
191192
}
192193

193194
self.experts = create_moe(**moe_params)

‎tests/integration/defs/.test_durations‎

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -273,18 +273,18 @@
273273
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-triton-auto]": 360.8670774899655953,
274274
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-trtllm-auto]": 360.00040231598541140556,
275275
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-trtllm-fp8]": 360.0003254589391872287,
276-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]": 745.8583740849863,
277-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]": 745.9345730679342523,
278-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]": 745.0004936959594488144,
279-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]": 745.00031642295653000474,
280-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]": 658.1757711600512,
281-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]": 745.9436021829606034,
282-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]": 745.0004371170070953667,
283-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]": 745.0004142870311625302,
284-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]": 676.3980704760179,
285-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]": 745.0292645250447094,
286-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]": 745.0003769229515455663,
287-
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]": 677.000331886054482311,
276+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]": 745.8583740849863,
277+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]": 745.9345730679342523,
278+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]": 745.0004936959594488144,
279+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]": 745.00031642295653000474,
280+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]": 658.1757711600512,
281+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]": 745.9436021829606034,
282+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]": 745.0004371170070953667,
283+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]": 745.0004142870311625302,
284+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]": 676.3980704760179,
285+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]": 745.0292645250447094,
286+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]": 745.0003769229515455663,
287+
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]": 677.000331886054482311,
288288
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]": 643.3513998010312,
289289
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[triton-auto]": 764.9216735750087537,
290290
"accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[trtllm-auto]": 764.0002969659981317818,

‎tests/integration/defs/accuracy/test_llm_api_pytorch.py‎

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4107,6 +4107,8 @@ def test_dummy_load_format(self):
41074107
task.evaluate(llm, is_integration_test=True)
41084108

41094109
@pytest.mark.skip_less_device(4)
4110+
@pytest.mark.parametrize("torch_compile", [False, True],
4111+
ids=["eager", "torch_compile"])
41104112
@pytest.mark.parametrize(
41114113
"kv_cache_dtype",
41124114
["auto", pytest.param("fp8", marks=skip_pre_blackwell)])
@@ -4127,7 +4129,7 @@ def test_dummy_load_format(self):
41274129
if x == 0 else "enable_configurable_moe")
41284130
def test_w4_4gpus(self, kv_cache_dtype, moe_backend, tp_size, pp_size,
41294131
ep_size, attention_dp, cuda_graph, overlap_scheduler,
4130-
enable_configurable_moe, mocker):
4132+
enable_configurable_moe, torch_compile, mocker):
41314133
# Handle ENABLE_CONFIGURABLE_MOE environment variable
41324134
if enable_configurable_moe == 1 and moe_backend not in [
41334135
"TRTLLM", "CUTLASS"
@@ -4157,10 +4159,13 @@ def test_w4_4gpus(self, kv_cache_dtype, moe_backend, tp_size, pp_size,
41574159
mocker.patch.object(GPQADiamond, "MAX_OUTPUT_LEN", MAX_OUTPUT_LEN)
41584160
mocker.patch.object(GPQADiamond, "MAX_INPUT_LEN", MAX_INPUT_LEN)
41594161

4162+
torch_compile_config = _get_default_torch_compile_config(torch_compile)
4163+
41604164
pytorch_config = dict(
41614165
disable_overlap_scheduler=not overlap_scheduler,
41624166
cuda_graph_config=CudaGraphConfig() if cuda_graph else None,
4163-
moe_config=MoeConfig(backend=moe_backend))
4167+
moe_config=MoeConfig(backend=moe_backend),
4168+
torch_compile_config=torch_compile_config)
41644169

41654170
kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.7,
41664171
dtype=kv_cache_dtype)

‎tests/integration/test_lists/qa/llm_function_core.txt‎

Lines changed: 24 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -587,18 +587,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
587587
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
588588
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
589589
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
590-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
591-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
592-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
593-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
594-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
595-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
596-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
597-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
598-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
599-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
600-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
601-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
590+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
591+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
592+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
593+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
594+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
595+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
596+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
597+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
598+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
599+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
600+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
601+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
602+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
603+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
604+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
605+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
606+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
607+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
608+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
609+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
610+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
611+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
612+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
613+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
602614
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
603615
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
604616
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

‎tests/integration/test_lists/qa/llm_function_core_sanity.txt‎

Lines changed: 24 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -92,18 +92,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
9292
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
9393
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
9494
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
95-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
96-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
97-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
98-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
99-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
100-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
101-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
102-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
103-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
104-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
105-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
106-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
95+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
96+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
97+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
98+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
99+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
100+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
101+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
102+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
103+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
104+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
105+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
106+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
107+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
108+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
109+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
110+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
111+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
112+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
113+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
114+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
115+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
116+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
117+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
118+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
107119
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
108120
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
109121
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

‎tests/integration/test_lists/qa/llm_function_nim.txt‎

Lines changed: 24 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -177,18 +177,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
177177
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
178178
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
179179
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
180-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
181-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
182-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
183-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
184-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
185-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
186-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
187-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
188-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
189-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
190-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
191-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
180+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
181+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
182+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
183+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
184+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
185+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
186+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
187+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
188+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
189+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
190+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
191+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
192+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
193+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
194+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
195+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
196+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
197+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
198+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
199+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
200+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
201+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
202+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
203+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
192204
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
193205
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
194206
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

‎tests/integration/test_lists/qa/llm_function_rtx6k.txt‎

Lines changed: 6 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -146,9 +146,12 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[True-True-cutlass-aut
146146
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-cutlass-auto]
147147
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[ep2-cutlass-auto]
148148
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
149-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
150-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
151-
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
149+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
150+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
151+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
152+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
153+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
154+
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
152155
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
153156
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]
154157
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm]

‎tests/integration/test_lists/test-db/l0_dgx_b200.yml‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -181,12 +181,12 @@ l0_dgx_b200:
181181
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8_chunked_prefill[tp4ep4-cuda_graph=True]
182182
- accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]
183183
- accuracy/test_disaggregated_serving.py::TestQwen3_30B_A3B::test_mixed_ctx_gen_model[ctxpp2gentp2]
184-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
185-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
186-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
187-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
188-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
189-
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
184+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
185+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
186+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
187+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
188+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
189+
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
190190
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-one_model-overlap_scheduler]
191191
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-one_model-no_overlap_scheduler]
192192
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-two_model-no_overlap_scheduler]

0 commit comments

Comments
 (0)