diff --git a/vllm/compilation/backends.py b/vllm/compilation/backends.py index cdae42fe4fce8..089d415ab5fb8 100644 --- a/vllm/compilation/backends.py +++ b/vllm/compilation/backends.py @@ -399,6 +399,7 @@ class VllmBackend: rank = vllm_config.parallel_config.rank dp_rank = vllm_config.parallel_config.data_parallel_rank local_cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}") + os.makedirs(local_cache_dir, exist_ok=True) self.compilation_config.local_cache_dir = local_cache_dir disable_cache = envs.VLLM_DISABLE_COMPILE_CACHE