[Kernel] changing fused moe kernel chunk size default to 32k (#7995)

vllm-project · Aug 30, 2024 · 34a0e96 · 34a0e96
1 parent 80c7b08
commit 34a0e96
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/vllm/envs.py b/vllm/envs.py
@@ -352,7 +352,7 @@ 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", "65536")),
+    lambda: int(os.getenv("VLLM_FUSED_MOE_CHUNK_SIZE", "32768")),
 
     # If set, vllm will skip the deprecation warnings.
     "VLLM_NO_DEPRECATION_WARNING":