[Bug] Fix batch invariance nvfp4 support (#39820)

Signed-off-by: yewentao256 <zhyanwentao@126.com>

.buildkite/test_areas/misc.yaml

@@ -224,6 +224,7 @@ steps:
     - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
     - VLLM_TEST_MODEL=deepseek-ai/DeepSeek-V2-Lite-Chat pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[TRITON_MLA]
     - VLLM_TEST_MODEL=Qwen/Qwen3-30B-A3B-Thinking-2507-FP8 pytest -v -s v1/determinism/test_batch_invariance.py::test_v1_generation_is_deterministic_across_batch_sizes_with_needle[FLASH_ATTN]
+    - pytest -v -s v1/determinism/test_nvfp4_batch_invariant.py
   
 - label: Acceptance Length Test (Large Models) # optional
   timeout_in_minutes: 25

vllm/model_executor/kernels/linear/__init__.py

@@ -601,7 +601,13 @@ def init_nvfp4_linear_kernel() -> NvFp4LinearKernel:
 
     # Env-var overrides.
     force_kernel: type[NvFp4LinearKernel] | None = None
-    if envs.VLLM_USE_FBGEMM:
+    if envs.VLLM_BATCH_INVARIANT:
+        logger.info_once(
+            "VLLM_BATCH_INVARIANT forces NVFP4 linear to use the "
+            "emulation backend for deterministic execution."
+        )
+        force_kernel = EmulationNvFp4LinearKernel
+    elif envs.VLLM_USE_FBGEMM:
         force_kernel = FbgemmNvFp4LinearKernel
     elif envs.VLLM_USE_NVFP4_CT_EMULATIONS:
         force_kernel = EmulationNvFp4LinearKernel