[compile] Nest inductor cache under AOT compile dir (#39718)
Signed-off-by: Alessandro Sangiorgi <asangior@redhat.com>
This commit is contained in:
@@ -507,6 +507,16 @@ def _support_torch_compile(
|
|||||||
hash_key,
|
hash_key,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
# Hash-level dir; shared across ranks on the same node.
|
||||||
|
self.compilation_config.local_cache_dir = cache_dir
|
||||||
|
inductor_cache = os.path.join(cache_dir, "inductor_cache")
|
||||||
|
os.makedirs(inductor_cache, exist_ok=True)
|
||||||
|
# Process-wide: post-load execution, CUDA-graph capture, and later
|
||||||
|
# autotune/recompile all need to write under {hash}/inductor_cache/.
|
||||||
|
# Unconditional because torch's cache_dir() may have pre-filled the
|
||||||
|
# /tmp default during import, making setdefault a no-op.
|
||||||
|
os.environ["TORCHINDUCTOR_CACHE_DIR"] = inductor_cache
|
||||||
|
|
||||||
rank = self.vllm_config.parallel_config.rank
|
rank = self.vllm_config.parallel_config.rank
|
||||||
dp_rank = self.vllm_config.parallel_config.data_parallel_index
|
dp_rank = self.vllm_config.parallel_config.data_parallel_index
|
||||||
cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}")
|
cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}")
|
||||||
|
|||||||
Reference in New Issue
Block a user