[UX] Add --moe-backend arg for explicit kernel selection (#33807)

Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>

tests/evals/gsm8k/configs/moe-refactor/Llama-4-Scout-BF16-fi-cutlass.yaml

@@ -2,8 +2,4 @@ model_name: "meta-llama/Llama-4-Scout-17B-16E-Instruct"
 accuracy_threshold: 0.92
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP16: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
-
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Llama-4-Scout-Fp8-ModelOpt-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/Llama-4-Scout-17B-16E-Instruct-FP8"
 accuracy_threshold: 0.92
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Llama-4-Scout-Fp8-ModelOpt-fi-trtllm.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/Llama-4-Scout-17B-16E-Instruct-FP8"
 accuracy_threshold: 0.92
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "latency"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_trtllm"

tests/evals/gsm8k/configs/moe-refactor/Llama-4-Scout-Fp8-ModelOpt-triton.yaml

@@ -2,6 +2,4 @@ model_name: "nvidia/Llama-4-Scout-17B-16E-Instruct-FP8"
 accuracy_threshold: 0.92
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "0"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=triton"

tests/evals/gsm8k/configs/moe-refactor/Mixtral-8x7B-BF16-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "mistralai/Mixtral-8x7B-v0.1"
 accuracy_threshold: 0.58
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP16: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Mixtral-8x7B-Fp8-AutoFp8-fi-cutlass.yaml

@@ -3,7 +3,4 @@
 # accuracy_threshold: 0.62
 # num_questions: 1319
 # num_fewshot: 5
-# server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-# env:
-#   VLLM_USE_FLASHINFER_MOE_FP8: "1"
-#   VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+# server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Nemotron-Nano-30B-Fp8-ModelOpt-fi-trtllm.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"
 accuracy_threshold: 0.29
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "latency"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_trtllm"

tests/evals/gsm8k/configs/moe-refactor/Nemotron-Nano-30B-NvFp4-ModelOpt-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4"
 accuracy_threshold: 0.29
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-BF16-fi-cutlass.yaml

@@ -2,6 +2,4 @@ model_name: "Qwen/Qwen3-30B-A3B"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP16: "1"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --enable-expert-parallel --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-Fp8-AutoFp8-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-Fp8-AutoFp8-fi-trtllm.yaml

@@ -2,7 +2,4 @@ model_name: "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "latency"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_trtllm"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-Fp8-AutoFp8-triton.yaml

@@ -2,7 +2,6 @@ model_name: "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=triton"
 env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "0"
   VLLM_USE_DEEP_GEMM: "0"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-Fp8-CT-Block-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "RedHatAI/Qwen3-30B-A3B-FP8-block"
 accuracy_threshold: 0.85
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-Fp8-CT-Block-triton.yaml

@@ -2,7 +2,6 @@ model_name: "RedHatAI/Qwen3-30B-A3B-FP8-block"
 accuracy_threshold: 0.85
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=triton"
 env:
-  VLLM_USE_FLASHINFER_MOE_FP8: "0"
   VLLM_USE_DEEP_GEMM: "0"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-CT-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "RedHatAI/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-CT-fi-trtllm.yaml

@@ -2,7 +2,4 @@ model_name: "RedHatAI/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "latency"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_trtllm"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-CT-vllm-cutlass.yaml

@@ -2,6 +2,4 @@ model_name: "RedHatAI/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "0"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-ModelOpt-fi-cutlass.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "throughput"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_cutlass"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-ModelOpt-fi-trtllm.yaml

@@ -2,7 +2,4 @@ model_name: "nvidia/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "1"
-  VLLM_FLASHINFER_MOE_BACKEND: "latency"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=flashinfer_trtllm"

tests/evals/gsm8k/configs/moe-refactor/Qwen3-30B-A3B-NvFp4-ModelOpt-vllm-cutlass.yaml

@@ -2,6 +2,4 @@ model_name: "nvidia/Qwen3-30B-A3B-NVFP4"
 accuracy_threshold: 0.88
 num_questions: 1319
 num_fewshot: 5
-server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2"
-env:
-  VLLM_USE_FLASHINFER_MOE_FP4: "0"
+server_args: "--enforce-eager --max-model-len 8192 --tensor-parallel-size 2 --moe-backend=cutlass"