[Bug] Gemma 4 31B crashes on k_eq_v full-attention layers (QKV split shape mismatch)

April 29, 2026 · #41283
View on GitHub
Python Difficulty: Medium

Parent Repository

vllm-project/vllm

Python repository

78,615 16,258
All Issues Back to vllm

Sign in required

Authenticate to use favourites & bookmarks

5