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

[bug-fix] Stabilize GLM-5.2 MTP IndexShare across PD and CUDA graph replay (#30839)

Browse Source 
Co-authored-by: kpham-sgl <khoa.pham@radixark.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Yuxuan Zhang
2026-07-13 19:37:07 -07:00
committed by GitHub
co-authored by kpham-sgl Claude Fable 5
parent 41ad0d9c26
commit 78dc581518
19 changed files with 371 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/debug_utils/pr_fix_toggle.py
+1 -5
View File
@@ -27,12 +27,8 @@ patches:
- target: sglang.srt.speculative.eagle_draft_cuda_graph_runner.EAGLEDraftCudaGraphRunner.capture_one_shape
edits:
- match: |
forward_batch.spec_info.hidden_states = hidden_states_backup
forward_batch.positions.sub_(self.eagle_worker.speculative_num_steps - 1)
return ret
replacement: |
forward_batch.spec_info.hidden_states = hidden_states_backup
return ret
replacement: ""
"""