Updated Nemotron recipes to use VLLM CUTLASS

2026-03-22 23:03:24 -07:00
parent 2d749742e4
commit 9e089acf2b
2 changed files with 5 additions and 14 deletions
--- a/recipes/nemotron-3-nano-nvfp4.yaml
+++ b/recipes/nemotron-3-nano-nvfp4.yaml
@@ -27,15 +27,10 @@ defaults:
  gpu_memory_utilization: 0.7
  max_model_len: 262144

-# Environment variables
-env:
-  VLLM_NVFP4_GEMM_BACKEND: "marlin"
-  VLLM_TEST_FORCE_FP8_MARLIN: "1"
-  VLLM_MARLIN_USE_ATOMIC_ADD: "1"
-
 # The vLLM serve command template
 command: |
  vllm serve nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4  \
+     --moe-backend cutlass \
     --max-model-len {max_model_len} \
     --port {port} --host {host} \
     --trust-remote-code \