minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Confine req-pool-idx assignment to the pool allocator (#25726)

Browse Source
This commit is contained in:
fzyzcjy
2026-05-19 09:22:51 +08:00
committed by GitHub
parent 6bcbf4de35
commit 5067da3eb6
1 changed files with 1 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/schedule_batch.py
+1 -4
View File
@@ -1774,9 +1774,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
self.extend_num_tokens = extend_num_tokens self.extend_num_tokens = extend_num_tokens
# Allocate memory # Allocate memory
out_cache_loc, req_pool_indices_tensor, req_pool_indices = alloc_for_extend( out_cache_loc, req_pool_indices_tensor, _ = alloc_for_extend(self)
self
)
# Set fields # Set fields
input_embeds = [] input_embeds = []
@@ -1792,7 +1790,6 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
mamba_track_seqlens_cpu = [] mamba_track_seqlens_cpu = []
for i, (req, seq_len, pre_len) in enumerate(zip(reqs, seq_lens, prefix_lens)): for i, (req, seq_len, pre_len) in enumerate(zip(reqs, seq_lens, prefix_lens)):
req.req_pool_idx = req_pool_indices[i]
assert seq_len - pre_len == req.extend_input_len assert seq_len - pre_len == req.extend_input_len
req.extend_batch_idx += 1 req.extend_batch_idx += 1