Update envs.py

Author: xiao-llm

vllm/envs.py

@@ -1181,8 +1181,8 @@ def get_vllm_port() -> Optional[int]:
 <<<<<<< HEAD
 =======
     # If set, use the fp8 mfma in rocm paged attention.
-    "VLLM_ROCM_USE_FP8_PAGED_ATTN":
-    lambda: bool(int(os.getenv("VLLM_USE_FP8_MFMA", "0"))),
+    "VLLM_ROCM_FP8_MFMA_PAGE_ATTN":
+    lambda: bool(int(os.getenv("VLLM_ROCM_FP8_MFMA_PAGE_ATTN", "0"))),
 
 >>>>>>> c588bf50e (Update vllm/envs.py)
     # Whether to use pytorch symmetric memory for allreduce