[Bug]: Ngram speculative decoding produces corrupted output on hybrid GDN (Qwen3.5) models

April 8, 2026 · #39273
View on GitHub
Python Difficulty: Easy

Parent Repository

vllm-project/vllm

Python repository

75,721 15,332
All Issues Back to vllm

Sign in required

Authenticate to use favourites & bookmarks

5