[None][feat] GPT_OSS torch compile support

tensorrt_llm/_torch/custom_ops/cpp_custom_ops.py

@@ -842,3 +842,24 @@ def _(
         # This is a fake implementation for shape inference
         # The actual operation modifies fused_q and q_pe in-place
         return None
+
+    @torch.library.register_fake("trtllm::mxfp8_quantize")
+    def _(input: torch.Tensor,
+          isSfSwizzledLayout: bool = True,
+          alignment: int = 32):
+        output_shape = list(input.shape)
+        padded_k = fp4_utils.pad_up(output_shape[-1], alignment)
+        output_shape[-1] = padded_k
+        output = input.new_empty(output_shape, dtype=torch.float8_e4m3fn)
+
+        sf_vec_size = 32
+        m = 1
+        for i in range(len(output_shape) - 1):
+            m *= output_shape[i]
+        if isSfSwizzledLayout:
+            sf_size = fp4_utils.pad_up(m, 128) * fp4_utils.pad_up(
+                padded_k // sf_vec_size, 4)
+        else:
+            sf_size = m * (padded_k // sf_vec_size)
+        sf = input.new_empty((sf_size, ), dtype=torch.uint8)
+        return output, sf

tensorrt_llm/_torch/models/modeling_gpt_oss.py

@@ -187,7 +187,8 @@ def __init__(
             'bias': True,
             'swiglu_alpha': self.swiglu_alpha,
             'swiglu_beta': self.swiglu_beta,
-            'swiglu_limit': self.swiglu_limit
+            'swiglu_limit': self.swiglu_limit,
+            'layer_idx': self.layer_idx,
         }
 
         self.experts = create_moe(**moe_params)

tests/integration/defs/.test_durations

@@ -273,18 +273,18 @@
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-triton-auto]": 360.8670774899655953,
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-trtllm-auto]": 360.00040231598541140556,
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-trtllm-fp8]": 360.0003254589391872287,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]": 745.8583740849863,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]": 745.9345730679342523,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]": 745.0004936959594488144,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]": 745.00031642295653000474,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]": 658.1757711600512,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]": 745.9436021829606034,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]": 745.0004371170070953667,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]": 745.0004142870311625302,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]": 676.3980704760179,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]": 745.0292645250447094,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]": 745.0003769229515455663,
-    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]": 677.000331886054482311,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]": 745.8583740849863,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]": 745.9345730679342523,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]": 745.0004936959594488144,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]": 745.00031642295653000474,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]": 658.1757711600512,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]": 745.9436021829606034,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]": 745.0004371170070953667,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]": 745.0004142870311625302,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]": 676.3980704760179,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]": 745.0292645250447094,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]": 745.0003769229515455663,
+    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]": 677.000331886054482311,
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]": 643.3513998010312,
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[triton-auto]": 764.9216735750087537,
     "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[trtllm-auto]": 764.0002969659981317818,

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -4107,6 +4107,8 @@ def test_dummy_load_format(self):
             task.evaluate(llm, is_integration_test=True)
 
     @pytest.mark.skip_less_device(4)
+    @pytest.mark.parametrize("torch_compile", [False, True],
+                             ids=["eager", "torch_compile"])
     @pytest.mark.parametrize(
         "kv_cache_dtype",
         ["auto", pytest.param("fp8", marks=skip_pre_blackwell)])
@@ -4127,7 +4129,7 @@ def test_dummy_load_format(self):
                              if x == 0 else "enable_configurable_moe")
     def test_w4_4gpus(self, kv_cache_dtype, moe_backend, tp_size, pp_size,
                       ep_size, attention_dp, cuda_graph, overlap_scheduler,
-                      enable_configurable_moe, mocker):
+                      enable_configurable_moe, torch_compile, mocker):
         # Handle ENABLE_CONFIGURABLE_MOE environment variable
         if enable_configurable_moe == 1 and moe_backend not in [
                 "TRTLLM", "CUTLASS"
@@ -4157,10 +4159,13 @@ def test_w4_4gpus(self, kv_cache_dtype, moe_backend, tp_size, pp_size,
         mocker.patch.object(GPQADiamond, "MAX_OUTPUT_LEN", MAX_OUTPUT_LEN)
         mocker.patch.object(GPQADiamond, "MAX_INPUT_LEN", MAX_INPUT_LEN)
 
+        torch_compile_config = _get_default_torch_compile_config(torch_compile)
+
         pytorch_config = dict(
             disable_overlap_scheduler=not overlap_scheduler,
             cuda_graph_config=CudaGraphConfig() if cuda_graph else None,
-            moe_config=MoeConfig(backend=moe_backend))
+            moe_config=MoeConfig(backend=moe_backend),
+            torch_compile_config=torch_compile_config)
 
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.7,
                                         dtype=kv_cache_dtype)

tests/integration/test_lists/qa/llm_function_core.txt

@@ -168,18 +168,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

tests/integration/test_lists/qa/llm_function_core_sanity.txt

@@ -114,18 +114,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

tests/integration/test_lists/qa/llm_function_nim.txt

@@ -162,18 +162,30 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-trtllm-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-fp8-torch_compile]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]

tests/integration/test_lists/qa/llm_function_rtx6k.txt

@@ -146,9 +146,12 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[True-True-cutlass-aut
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[tp2-cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[ep2-cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_2gpus[dp2-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto-torch_compile]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
+accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-torch_compile]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto]
 accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm]

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

@@ -182,12 +182,12 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8_chunked_prefill[tp4ep4-cuda_graph=True]
   - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]
   - accuracy/test_disaggregated_serving.py::TestQwen3_30B_A3B::test_mixed_ctx_gen_model[ctxpp2gentp2]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto]
-  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass-auto-eager]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-triton-auto-eager]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-auto-eager]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-trtllm-fp8-eager]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto-eager]
+  - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-triton-auto-eager]
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-one_model-overlap_scheduler]
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-one_model-no_overlap_scheduler]
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[trtllm-two_model-no_overlap_scheduler]