minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Confine max-prefix-len to where it is used and drop the leftover variable (#25719)

Browse Source
This commit is contained in:
fzyzcjy
2026-05-19 09:20:21 +08:00
committed by GitHub
parent 1cba3ab467
commit 32f1259c91
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/schedule_batch.py
+1 -1
View File
@@ -1022,11 +1022,11 @@ class Req(ReqDllmMixin):
max_prefix_len = min(max_prefix_len, self.logprob_start_len) max_prefix_len = min(max_prefix_len, self.logprob_start_len)
max_prefix_len = max(max_prefix_len, 0) max_prefix_len = max(max_prefix_len, 0)
token_ids = self.fill_ids[:max_prefix_len] token_ids = self.fill_ids[:max_prefix_len]
del max_prefix_len
# Disable prefix caching when embed overrides are present: same token IDs # Disable prefix caching when embed overrides are present: same token IDs
# with different override vectors must not share cached KV values. # with different override vectors must not share cached KV values.
if self.positional_embed_overrides is not None: if self.positional_embed_overrides is not None:
max_prefix_len = 0
token_ids = [] token_ids = []
if tree_cache is not None: if tree_cache is not None: