Back to AIBriefs
AnalysisAI Models

Gemma4 31B comparison of Q4_K_M, QAT, heretic quantizations

User shares experience running Gemma4 31B with different quantizations, noting the UD Q4_K_M version as a 'functional nervous wreck' due to hyper-vigilant behavior. The heretic version is used as a break from the overly cautious default.

·
8 days ago
Gemma4 31B comparison of Q4_K_M, QAT, heretic quantizations — AIBriefs