b10539 b10538
ggml-org/llama.cppAug 21, 2026 → Aug 21, 20260 days between releases
Generating AI comparison
Analyzing what changed between b10539 and b10538. This page updates automatically — usually within a minute.
The Two Releases
b10539
b10539
Aug 21, 2026· 2 months ago
Fixes a potential overflow in the Vulkan Flash Attention (FA) MMQ kernel where Q quantization calculations used denormalized floats.
Full release notesb10538
b10538
Aug 21, 2026· 2 months ago
Optimizes the macOS Metal backend by dequantizing the Key-Value (KV) cache only for large batches to reduce overhead and improve efficiency.
Full release notes