ggml-org/llama.cpp

Pull Requests Commits

add --kv-shift to CI

ngxson committed 1 day ago

acba82b7

ngxson committed 1 day ago

b76df217

update review skill

ngxson committed 1 day ago

e438928b

migrate new models

ngxson committed 1 day ago

0b91231b

Merge branch 'master' into xsn/fix_kv_shift

ngxson committed 1 day ago

a1f99a4e

llama: add graph build_rope (#31693)

ngxson committed 1 day ago

Verified f7d2a2f6

model : add Ripple (ricursiveai) (#31692)

notforu777 committed 1 day ago

Verified fecdc81e

model: Add support for the xing4_0 (#31589)

shuxiaoqiong committed 1 day ago

Verified 45b6604b

ui: prevent a hidden sidebar from intercepting clicks (#31712)

ntowle committed 1 day ago

Verified e30ad80f

Merge branch 'master' into xsn/fix_kv_shift

ngxson committed 1 day ago

685e8b3c

llama: refactoring: move model to class member of llm_graph_context (#31733)

ngxson committed 1 day ago

Verified e0f96a1a

server : simplify windows MCP command resolution (#31728)

angt committed 1 day ago

Verified 468fde90

vendor : update cpp-httplib to 0.65.0 (#31731)

angt committed 1 day ago

Verified 1416f76a

ggml-cuda: sanitize post-bias NaNs in MoE selection (#31005)

ArberSephirotheca committed 1 day ago

Verified 10d4846a

vulkan: Implement test-fusion interface (#30769)

jeffbolznv committed 1 day ago

Verified 61d44723

CUDA: use highest compiled arch in get_mmvq_mmid_max_batch (#31679)

ManuelPDM committed 1 day ago

Verified 016c1d91

vulkan: RDNA3/4 Large-Tile Split-K Optimization (#30280)

Alex-JP-93 committed 1 day ago

Verified da3416d8

gguf-py : external chat templates have precedence over tokenizer_config (#31717)

CISC committed 1 day ago

Verified cc78da97

metal : fuse DeepSeek-V4/V4.1 and GLM MoE routing (#31615)

ggerganov committed 1 day ago

Verified ea40a3a1

CUDA/HIP: matrix-core (MFMA) lightning indexer for CDNA2 (gfx90a) (#29050)

jadenmach2 committed 1 day ago

Verified 15e50990

CUDA: linearize MoE data layout for MMQ/MMF (#31047)

JohannesGaessler committed 1 day ago

Verified 69ee2ecc

vulkan: FWHT kernels for block widths above 512 (#29772)

bri-prism committed 1 day ago

Verified 2bc46baa

vulkan : check combined K+V f16 scratch size in FA dequant path (#30487)

nisbenz committed 1 day ago

Verified 3b83d12b

llama-server: Use reference-counted checkpoints instead of deep-copies (#30137)

cwriter committed 1 day ago

Verified e152332a

ggml : support `truncate=false` for YaRN (#30206)

tokinasin committed 1 day ago

Verified 1fa7f1c4

conversion : resolve quantized_layers from source tensor names (#31611)

danbev committed 1 day ago

Verified 9c9074e1

metal : use mul_mm for 32 <= K < 64 (#30982)

vieduy committed 1 day ago

Verified f7b41cca

server : reject non-object session updates (#31714)

angt committed 1 day ago

Verified c55b0948

fit : account for skipped NextN layers when setting n_gpu_layers (#31569)

am17an committed 1 day ago

Verified 2c192444

webgpu : add MEAN support (#31591)

LeON-Nie-code committed 1 day ago

Verified 553d2c2e

Older