ggerganov/llama.cpp

Pull Requests Commits

qwen4exp: fix sparse-attention block selection

tarruda committed 24 days ago

Verified f91123d2

metal : add fa-vec tunings for M4 (#27875)

Strongtut committed 25 days ago

Verified f5e85d43

OpenVINO: Update OV to 2026.3.1, whisper.cpp support, Qwen3.5 on NPU, and new ops (#27843)

wine99 committed 25 days ago

Verified 511f9c13

convert: prevent ndarray conversion in LazyChunkedTensor (#27869)

ngxson committed 25 days ago

Verified b19cbe92

sycl: use TILE for quantized KV decode on BMG (#26689)

johnkarlhill committed 25 days ago

Verified d077b4c2

sycl: bind the f16 KV cache in place for the oneDNN SDPA path (#27468)

Titaniumtown committed 25 days ago

Verified be876204

metal : add fa-vec tunings for M3 Max, M5 and M5 Pro (#27863)

ggerganov committed 25 days ago

Verified 8963a9bd

metal : add fa-vec tunings for M4 Pro (#27824)

infinitewarp committed 25 days ago

Verified 6d6b697c

tests : run test-save-load-state across all architectures (#27755)

ggerganov committed 25 days ago

Verified 4e97ac86

model: add DSpark support for Nemotron3.5 (#27804)

ruixiang63 committed 25 days ago

Verified ca3d5a3e

ggml-hexagon: add HTP unary ops for ABS and LOG (#27786)

cqderek committed 25 days ago

Verified e70802a0

hex-unary: fix RMS_NORM_MUL weight-offset bugs for grouped/broadcast norms (#27798)

aparmp-quic committed 25 days ago

Verified 83d855c5

server: add ctx-per-slot (--kv-unified-per-slot) (#24124)

bartowski1182 committed 25 days ago

Verified 18443257

ci : build only the ggml-hip backend for windows-rocm release (#27753)

harkgill-amd committed 25 days ago

Verified 32176338

model: add Qwen3.8-Flash-Next (qwen4exp) (#27742)

danielhanchen committed 25 days ago

Verified 6c84c7d5

ci : bundle HIP runtime DLLs with Windows ROCm release (#26973)

slojosic-amd committed 25 days ago

Verified 6fdd0ac8

spec : add DFlash2 support (local convolution + candidate selector) (#27342) (#27816)

ngxson committed 25 days ago

Verified b10f9ca5

opencl: add bin kernels `kernel_gemm_moe_q4_0_q8_1_dp4a_bin`, `kernel_gemm_moe_mxfp4_q8_1_dp4a_bin` (#27768)

shawngu-quic committed 25 days ago

Verified 58546250

quantize: cap working memory size to avoid loading big tensors onto RAM (#27795)

ngxson committed 25 days ago

Verified 732707df

Feature: Added LIGHTNING_INDEXER support for Deepseek V4 ops on Vulkan Backend (#27453)

shenron0101 committed 26 days ago

Verified cb300598

pr2wt : use ssh/https remote in worktree depending on base (#27800)

CISC committed 26 days ago

Verified 1a946ec7

llama: model_loader: add TENSOR_READ_LAZY (#27794)

ngxson committed 26 days ago

Verified fac889fb

ui: Improve Chat Form Actions UI/UX (models selector, add panel) (#27746)

allozaur committed 26 days ago

Verified cae63579

convert : fix Nemotron-H LoRA GGUF conversion (#27356)

frozenblade1224 committed 26 days ago

Verified bcb6084a

ui: Replace per-conversation MCP overrides with per-conversation tool policy (#27745)

allozaur committed 26 days ago

Verified fe235f43

spec: Add benchmark-only synthetic speculative acceptance options (#27711)

gaugarg-nv committed 26 days ago

Verified 2bb9bdda

model : simplify MiniMax-01 graph (#27790)

fairydreaming committed 26 days ago

Verified deae5ee1

args: add --video-* CLI arguments (#24318)

ngxson committed 26 days ago

Verified f2955121

metal : fix memory leaks due to missing autoreleasepools (#27758)

nikwen committed 26 days ago

Verified 915dc6d3

llama : add --n-cpu-ffn option (#26622)

John-194 committed 26 days ago

Verified c5fc7e34

Older