Revert "[Kernel] changing fused moe kernel chunk size default to 32k (v…

…llm-project#7995)" (opendatahub-io#207) This reverts commit 34a0e96.
prarit · Sep 25, 2024 · cc2039c · cc2039c
1 parent 48c0cb4
commit cc2039c
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/vllm/envs.py b/vllm/envs.py
@@ -404,7 +404,7 @@ def get_default_config_root():
             os.path.join(get_default_cache_root(), "vllm", "xla_cache"),
         )),
     "VLLM_FUSED_MOE_CHUNK_SIZE":
-    lambda: int(os.getenv("VLLM_FUSED_MOE_CHUNK_SIZE", "32768")),
+    lambda: int(os.getenv("VLLM_FUSED_MOE_CHUNK_SIZE", "65536")),
 
     # If set, vllm will skip the deprecation warnings.
     "VLLM_NO_DEPRECATION_WARNING":