chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Fix] data race in req_to_token pool (#17850)

Browse Source
This commit is contained in:
cctry
2026-02-02 14:38:15 -08:00
committed by GitHub
parent cbf1500390
commit 027f314050
13 changed files with 107 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/mem_cache/radix_cache.py
View File

@@ -451,7 +451,6 @@ class RadixCache(BasePrefixCache):
req.req_pool_idx, :kv_committed_len
]
self.token_to_kv_pool_allocator.free(kv_indices)
self.req_to_token_pool.free(req.req_pool_idx)
return
token_ids = (req.origin_input_ids + req.output_ids)[:kv_committed_len]
@@ -485,7 +484,6 @@ class RadixCache(BasePrefixCache):
self.token_to_kv_pool_allocator.free(kv_indices[len(keys) :])
# Remove req slot release the cache lock
self.req_to_token_pool.free(req.req_pool_idx)
self.dec_lock_ref(req.last_node)
def cache_unfinished_req(self, req: Req, chunked=False):