Releases: KakaruHayate/game.cpp
Releases · KakaruHayate/game.cpp
Release list
v0.1.1
What's Changed
- perf: DBCache robustness knobs + bn-block experiment (edge-dit borrow) by @KakaruHayate in #13
Full Changelog: v0.1.0...v0.1.1
v0.1.0
What's Changed
- Merge quantization/dwconv/DBCache work into main by @KakaruHayate in #1
- ci: ship both full (F32) and quantized (Q8_0) artifacts by @KakaruHayate in #2
- fix: segmenter tail latent registered before gallocr alloc (DBCache crash) by @KakaruHayate in #3
- perf: enable DBCache by default for multi-step D3PM (nsteps>1) by @KakaruHayate in #4
- docs: backend × weights guidance from local EP benchmark by @KakaruHayate in #5
- chore: absorb ggml-metal-binary-archive into game.cpp (ecosystem dissolution) by @KakaruHayate in #6
- perf: EP-aware DBCache default (CPU 0.25 on, GPU off) by @KakaruHayate in #7
- ci: add linux-x64-cpu (GGML_NATIVE=true) package by @KakaruHayate in #8
- perf(Vulkan): persistent VkPipelineCache (cold-start fix) by @KakaruHayate in #9
- perf: dedicated dwconv kernel on Vulkan/Metal/CUDA (+11-19% Vulkan) by @KakaruHayate in #10
- docs: Windows+CUDA toolchain notes from a successful local CUDA build by @KakaruHayate in #11
- perf: fuse attn Q+KV into one qkv linear (llama-style) by @KakaruHayate in #12
Full Changelog: https://github.com/KakaruHayate/game.cpp/commits/v0.1.0