ggml-org/llama.cpp

Pull Requests Commits

appropriately flag noscan ssm_a tensors

CISC committed 6 hours ago

Verified e2f2a18a

CUDA: XOR swizzle flash attn K,V smem fp16 tiles (#25635)

ynankani committed 6 hours ago

Verified e4b9af00

metal : add concat support for quantized types (#28116)

ggerganov committed 6 hours ago

Verified ab0b3bd3

AVX2: Speed up large batch size prompt processing of IQ models (#27402)

bartowski1182 committed 8 hours ago

Verified 85c55223

metal : add top-k radix implementation (#28073)

ggerganov committed 8 hours ago

Verified 2a74817f

kv-cache : optimize restoring non-contiguous cells (#27991)

itsnotoger committed 10 hours ago

Verified 2d8d612e

opencl: tune the quant paths for Intel Xe-LP GPUs to improve its TG and PP performance (#26438)

wanghqc committed 10 hours ago

Verified 010be968

ui: copy the displayed text of grouped agentic responses (#27832)

ServeurpersoCom committed 11 hours ago

Verified 774ee0e2

webgpu : avoid crash when offset is not multiple of 4 in WebGPU ggml_backend_tensor_get() implementation (#28045)

fairydreaming committed 12 hours ago

Verified 8e53fcef

ROCm: add radix TOP_K for long rows (#27466)

jadenmach2 committed 13 hours ago

Verified f8dbcd61

metal : add fa-vec tunings for M1 (#28078)

nikwen committed 14 hours ago

Verified 5d4a3be2

CUDA: extend MOE fusion to specdec, earlier MOE glu fusion and topk-router fusion were restricted to 1 token (#27621)

ynankani committed 15 hours ago

Verified 41ef91f7

sycl : Enhance to get the free memory of Intel GPU (#27968)

arthw committed 16 hours ago

Verified a32af33d

ci : add check for unzip (#28082)

CISC committed 16 hours ago

Verified 580e88d8

spec : fuse the DFlash encoder into the KV cache injection (#27310)

wjinxu committed 17 hours ago

Verified 662a0b01

vulkan: tune mat-vec rows for batched inference on Strix Halo (#27909)

SimonTeixidor committed 17 hours ago

Verified 2cdae802

ggml : add MUL_MAT to the list of ops that may need additional memory (for WebGPU) (#28071)

fairydreaming committed 18 hours ago

Verified 557614e0

vulkan: top_k radix select for k >= 1024 for Qwen 3.8 Flash Next (#28032)

0cc4m committed 21 hours ago

Verified daef7b68

hexagon: fix CPY fence bug (#28033)

yshsharke committed 1 day ago

Verified 9723942a

metal : add remaining Q4_1/Q5_0/Q5_1 fa-vec tunings for M2 (#28017)

ring2003 committed 1 day ago

Verified bd55e6aa

rpc: avoid serializing buffers from other servers (#26500)

hmirin committed 1 day ago

Verified a7cc83bb

ggml : fix ggml_backend_buft_get_alloc_size() guard (#28038)

ggerganov committed 1 day ago

Verified 6d1479c1

kv-cells: stop the sequence scan once all sequences are seen (#28011)

ServeurpersoCom committed 1 day ago

Verified 62acc89c

ggml: add SWIGLU_CLAMP (#27930)

am17an committed 1 day ago

Verified 0190529e

llama: improve TENSOR_READ_LAZY handling (#27837)

ngxson committed 1 day ago

Verified 25781383

CUDA: use the fast mm_ids_helper path for any n_expert_used (#27978)

ServeurpersoCom committed 1 day ago

Verified f1793c1c

hip: tune rdna 3 mmq config (#26284)

itterative committed 1 day ago

Verified 0b5be7e4

hip : optimize Q2_0 dot-product path for gfx1201 (#26753)

LunalFresh committed 1 day ago

Verified e4221480

dflash: pass missing NVFP4 scales to attention operations (#28000)

JamePeng committed 1 day ago

Verified cc231cb0

common: rename --tensor-read-lazy to --lazy-mode, add -lzm shorthand (#27969)

ggerganov committed 1 day ago

Verified bebc9350

Older