ngxson/llama.cpp

Pull Requests Commits

mtmd: use causal attn for gemma 4 audio

ngxson committed 32 days ago

88956064

convert : force f16 or f32 on step3-vl conv weights (#21646)

CISC committed 32 days ago

Verified 1e9d771e

mtmd: add gemma 4 test (vision + audio) [no ci] (#21806)

ngxson committed 32 days ago

Verified aa4695c5

mtmd: add Gemma 4 audio conformer encoder support (#21421)

stephencox-ict committed 32 days ago

Verified 547765a9

fix: Proper messages rendering for "Show raw output" (#21672)

allozaur committed 32 days ago

Verified 9e209c5a

docs: add guide on how to add multimodal support (#21778)

ngxson committed 32 days ago

Verified 6313acbe

CUDA: skip compilation of superfluous FA kernels (#21768)

JohannesGaessler committed 33 days ago

Verified ff5ef827

mtmd : add MERaLiON-2 multimodal audio support (#21756)

SiruiHe committed 33 days ago

Verified 073bb2c2

opencl: add basic support for q5_k (#21593)

shaofeiqi committed 33 days ago

Verified af1127d3

TP: fix Qwen 3 Next data split (#21732)

JohannesGaessler committed 34 days ago

Verified 865ff06b

ggml : fix a few instances of missing GGML_TYPE_Q1_0 cases (#21716)

CISC committed 34 days ago

Verified 2b2cd57d

py : Bump typer to latest to fix huggingface_hub issue (#21701)

bartowski1182 committed 34 days ago

Verified 660386f6

CUDA: also store node->src ne/nb for graph equality (#21736)

am17an committed 34 days ago

Verified a29e4c0b

fix: Fix broken structured output when using $refs in json_schema (#21699)

Galunid committed 34 days ago

Verified b136b62c

hexagon: add support for linux on snapdragon (#21707)

tboinovski1 committed 34 days ago

Verified 81069a80

hexagon: improved Op queuing, buffer and cache management (#21705)

max-krasnyansky committed 34 days ago

Verified 9aa28077

common : better align to the updated official gemma4 template (#21704)

aldehir committed 34 days ago

Verified 3fc65063

common : add callback interface for download progress (#21735)

angt committed 34 days ago

Verified 05b3caaa

model : make Gemma 4 shared-KV tail attn_k tensors optional on load (#21739)

MoonRide303 committed 34 days ago

Verified e62fa13c

ggml-webgpu: support non-square subgroup matrix configs for Intel GPUs (#21669)

SharmaRithik committed 34 days ago

Verified bfd1f453

ggml-webgpu: address quantization precision and backend lifecycle managment (#21521)

Constannnnnt committed 34 days ago

Verified e4fed9d0

server : ignore --alias when using --models-preset (#21380)

angt committed 34 days ago

Verified 5dd10253

common : fix when loading a cached HF models with unavailable API (#21670)

angt committed 34 days ago

Verified fb38d6f2

common: mark --split-mode tensor as experimental (#21684)

JohannesGaessler committed 34 days ago

Verified 0893f50f

webui: Static build output improvements (#21667)

allozaur committed 34 days ago

Verified f989a6e3

common : enable reasoning budget sampler for gemma4 (#21697)

berkidem committed 34 days ago

Verified d7ff074c

docs : fix broken link to ggml-openvino in OPENVINO.md (#21709)

ibelem committed 34 days ago

Verified 3f8752b5

vulkan: Support Q1_0 (#21539)

jeffbolznv committed 35 days ago

Verified 7b691253

common : add fluidity to the progress bar (#21671)

angt committed 35 days ago

Verified e095a482

CUDA: fuse muls (#21665)

am17an committed 35 days ago

Verified e34f0421

Older