vllm-ascend vnpu v1

2025-12-26 07:37:35 +00:00
parent 2f1aed98cc
commit 135cc0a505
168 changed files with 28337 additions and 9 deletions
--- a/vllm_ascend/envs.py
+++ b/vllm_ascend/envs.py
@@ -166,7 +166,8 @@ env_variables: Dict[str, Callable[[], Any]] = {
    lambda: bool(int(os.getenv("VLLM_ASCEND_ENABLE_MLAPO", '0'))),
    # Whether to enable transpose weight and cast format to FRACTAL_NZ.
    "VLLM_ASCEND_ENABLE_NZ":
-    lambda: int(os.getenv("VLLM_ASCEND_ENABLE_NZ", 1)),
+    lambda: int(os.getenv("VLLM_ASCEND_ENABLE_NZ", 0)),
+    "VLLM_ASCEND_ENABLE_IDLE_OFFLOAD": lambda: int(os.getenv("VLLM_ASCEND_ENABLE_IDLE_OFFLOAD", 1)),
 }

 # end-env-vars-definition