python/sglang/srt/speculative/ngram_utils.py · 0b2aa8a70ca503abbde4905cf0f857933fbdc928 · zhaoyu6 / sglang · GitLab

Find file Blame History Permalink

Intoduce cpu tensor as metadata to avoid blocking gpu kernel launch (#10720) · 0b2aa8a7
Zhang Junda authored Oct 02, 2025
```
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
```
0b2aa8a7

ngram_utils.py 16 KB