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

[Bugfix] Add rids to the batch filtering for two batch overlap (#19418)

Browse Source
This commit is contained in:
Shangming Cai
2026-02-26 06:57:25 -08:00
committed by GitHub
parent 97f1fa5e6b
commit e55e65535e
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/batch_overlap/two_batch_overlap.py
+1
View File
@@ -653,6 +653,7 @@ class TboForwardBatchPreparer:
"extend_seq_lens_cpu", "extend_seq_lens_cpu",
"extend_logprob_start_lens_cpu", "extend_logprob_start_lens_cpu",
"lora_ids", "lora_ids",
"rids",
]: ]:
old_value = getattr(batch, key) old_value = getattr(batch, key)
if old_value is None: if old_value is None:
python/sglang/srt/server_args.py
+1 -1
View File
@@ -1048,7 +1048,7 @@ class ServerArgs:
reserved_mem += 6 * 1024 reserved_mem += 6 * 1024
elif self.speculative_algorithm != "NGRAM": elif self.speculative_algorithm != "NGRAM":
# eagle draft models and cuda graphs # eagle draft models and cuda graphs
reserved_mem += 2 * 1024 reserved_mem += 4 * 1024
self.mem_fraction_static = ( self.mem_fraction_static = (
round((gpu_mem - reserved_mem) / gpu_mem, 3) round((gpu_mem - reserved_mem) / gpu_mem, 3)