perf: turbo VEC flash attention — +9% decode on CUDA via autoresearch#53

Open

signalnine wants to merge 153 commits into

TheTom:feature/turboquant-kv-cachefrom

signalnine:pr/fattn-vec-turbo-opts

Commits on Apr 2, 2026

Commits on Apr 3, 2026

fix: remove redundant extern from GGML_API macro (GCC 13.3 hard error)

TheTom
and
claude
committed
Merge remote-tracking branch 'origin/master' into feature/turboquant-kv-cache
TheTom
committed

Commits on Apr 4, 2026

Enhance Metal operations for TQ weights and concurrency handling for Gemma 4
iamwavecut
committed
Update GGMLQuantizationType and LlamaFileType enums to include TQ3_1S and TQ4_1S quantization types with corresponding sizes in GGML_QUANT_SIZES.
iamwavecut
committed

Commits on Apr 6, 2026

Commits on Apr 8, 2026

Commits on Apr 9, 2026