GQA/MQA attention broken — only MHA (Q_heads == KV_heads) produces coherent output
April 12, 2026 · #61
c
Difficulty: Medium
Labels
bug
Parent Repository
quantumaikr/quant.cpp
c repository
377 43
Labels
quantumaikr/quant.cpp
c repository
Sign in required
Authenticate to use favourites & bookmarks