LLM inference in C/C++
C++ ggml
Commits on master
metal: fix NaN in mul_mm_id when activations exceed f16 range (#26223)
0a8b29a Michael de Gans committed 2 months ago
hexagon: add back missing contiguous fast-path and hvx_copy_uu for each run (#28886)
930e2fa Jhen-Jie Hong committed 2 months ago
hex-cpy: use dma if src and dst are contiguous (#28906)
72b590d Trivikram Reddy committed 2 months ago
opencl: choose the MoE expert matmul by batch size for speculative decoding/MTP (#27637)
9f31776 Hongqiang Wang committed 2 months ago
docs: Rule of thumb for AI review time [no ci] (#28945)
6011c34 Johannes Gäßler committed 2 months ago
models : move build_arch_graph() after graph() template specialization (#28934)
9e71716 Chris Peterson committed 2 months ago
OpenVINO: optimize stateful decode and GPU MoE inference (#28638)
77d554b Zijun Yu committed 2 months ago
metal : add FA kernels for HSK=96, HSV=64 (MiniCPM3) (#28599)
1e7bcf3 Yanzhao Wang committed 2 months ago
ci: Bump CUDA Windows x64 builds to 13.4.1 (#28930)
0ecb159 shivamkumard-ctrl committed 2 months ago
cmake : use PROJECT_SOURCE_DIR instead of CMAKE_SOURCE_DIR (#28771)
69eb250 Daniel Bevenius committed 2 months ago
webui: stop re-probing disabled /tools endpoint on every message (#28646)
1bc7a5a Abhiram committed 2 months ago
ci : reuse build tag name when used instead of safe one (#28911)
7cf1c54 Sigbjørn Skjæret committed 2 months ago
CI: hip-quality-check: ignore spill added in bfdc32183d57f1e35bacf35c47d6311e2028bbbc (#28909)
96ffdc4 uvos committed 2 months ago