vllm/spec_decode/spec_decode_worker.py · 9804145cac8b108725b2a431be6cd8a65d20ef9a · OpenDAS / vllm_cscc · GitLab

Find file Blame History Permalink

[Model][Speculative Decoding] Expand DeepSeek MTP code to support k > n_predict (#13626) · 9804145c
Benjamin Chislett authored Feb 27, 2025
```
Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai>
```
9804145c

spec_decode_worker.py 61.5 KB