You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
MLX-native port of KVarN — variance-normalized KV-cache quantization for Apple Silicon. 3.3× compression at 71% FP16 speed, matches FP16 on GSM8K within ~2%.