add tests

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

Author: syuoni

tensorrt_llm/_torch/modules/fused_moe/fused_moe_cute_dsl.py

@@ -90,7 +90,7 @@ def pad_and_multiply(scale, tensor):
     return ref
 
 
-def cute_dsl_nvfp4_group_gemm_ref(
+def cute_dsl_nvfp4_grouped_gemm_ref(
     a: torch.Tensor,
     b: torch.Tensor,
     a_sf: torch.Tensor,

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -1674,13 +1674,15 @@ def test_nvfp4_4gpus_online_eplb(self, fp8kv):
                            (False, False, False, True),
                            (True, False, True, True), (True, True, True, True)])
     @parametrize_with_ids("mtp_nextn", [0, 2])
-    @parametrize_with_ids("moe_backend", ["CUTLASS", "TRTLLM"])
+    @parametrize_with_ids("moe_backend", ["CUTLASS", "TRTLLM", "CUTEDSL"])
     def test_nvfp4(self, fp8kv, attention_dp, cuda_graph, overlap_scheduler,
                    torch_compile, mtp_nextn, 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")
+        if moe_backend == "CUTEDSL" and get_sm_version() != 100:
+            pytest.skip(f"{moe_backend} backend supports SM 100 only")
 
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.75)
         torch_compile_config = TorchCompileConfig(
@@ -1767,7 +1769,7 @@ def test_nvfp4_batch_waiting(self, torch_compile, fp8kv, cuda_graph,
                                                          (2, 2, 1), (1, 4, 1)],
                              ids=["tp4", "ep4", "tp2pp2", "pp4"])
     @parametrize_with_ids("mtp_nextn", [0, 2])
-    @parametrize_with_ids("moe_backend", ["CUTLASS", "TRTLLM"])
+    @parametrize_with_ids("moe_backend", ["CUTLASS", "TRTLLM", "CUTEDSL"])
     def test_nvfp4_4gpus(self, fp8kv, attention_dp, cuda_graph,
                          overlap_scheduler, tp_size, pp_size, ep_size,
                          torch_compile, mtp_nextn, moe_backend):
@@ -1777,6 +1779,9 @@ def test_nvfp4_4gpus(self, fp8kv, attention_dp, cuda_graph,
                                         or get_sm_version() == 121):
             pytest.skip(
                 "MOE TRTLLM backend does not support SM version 120 or 121")
+        if moe_backend == "CUTEDSL" and get_sm_version() != 100:
+            pytest.skip(f"{moe_backend} backend supports SM 100 only")
+
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.75)
         # Picewise Cuda Graph cannot be enabled for nvfp4 attention dp.
         torch_compile_config = TorchCompileConfig(

tests/integration/test_lists/test-db/l0_b200.yml

@@ -27,6 +27,7 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False] ISOLATION
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=nvfp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]
@@ -146,4 +147,5 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestSeedOss_36B::test_auto_dtype

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

@@ -38,6 +38,7 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp2pp2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]
@@ -167,6 +168,7 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-pp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=False]