ggerganov/llama.cpp

Pull Requests Commits

test-save-load-state : compare logits with NMSE and feed expected tokens

ggerganov committed 28 days ago

b9c61e05

vendor : update cpp-httplib to 0.57.0 (#29214)

angt committed 29 days ago

Verified 6f41ac59

docker : bump cuda to 13.4.1 (#29207)

CISC committed 29 days ago

Verified ec91ab5a

sycl : support gated DSV4_HC_PRE and optional HC_POST comb matrix (#29132)

cwriter committed 29 days ago

Verified bb3c853c

sycl : pinned memory use right device context instead of 0 (#28895)

lslusarczyk committed 29 days ago

Verified af911149

CUDA: Follow up of #25635, refactoring FA shared smem swizzle (#28536)

ynankani committed 29 days ago

Verified 1884824f

test-llama-archs : make tensor data stdev configurable and improve help (#29133)

ggerganov committed 29 days ago

Verified 161755f2

tests/test-backend-ops : allow regex entries in the -o filter (#29204)

ggerganov committed 29 days ago

Verified 1d72b05d

ci : refactor build-self-hosted into backend-specific workflows (#28991)

CISC committed 29 days ago

Verified 542e9202

args: add env vars for temperature, top-p, min-p and penalties (#27380)

kucharskim committed 29 days ago

Verified e0dff584

server : do not forward --api-key-file to router-spawned child instances (#28938)

nandan2003 committed 29 days ago

Verified 982a3329

tests : remove stale comment (#29140)

mostafafaheem committed 29 days ago

Verified 711f60be

ggml-metal : simplify fusion pattern op list declaration (#29206)

ggerganov committed 29 days ago

Verified 335b21fc

json: Fixed json enum handling (#28518)

Silverside committed 29 days ago

Verified 26394b4e

sycl : coalesce MKL-FA softmax loads instead of one work-item per row (#28918)

anantshri committed 29 days ago

Verified 1aa2954b

ggml-cpu: ARM Repack kernels for Q1_0 (#23492)

pl752 committed 29 days ago

Verified 8034c1d1

ci : Upgrade CUDA to 13.4 for Ubuntu CUDA Release Builds (#29202)

sam-india-007 committed 29 days ago

Verified 6ad1af56

hexagon: overhaul of buffer and DMA handling to support 64bit mappings + improvements (#29197)

max-krasnyansky committed 29 days ago

Verified 0c3626ec

cuda : tune MMVQ to MMQ crossover for SM70 (Volta) (#28912)

cyyself committed 29 days ago

Verified 68d9053a

metal : fix deprecation warnings from macOS 27 SDK (#29136)

nikwen committed 29 days ago

Verified 8aa161b5

webgpu : add fused gdn + cpy (#28976)

yomaytk committed 29 days ago

Verified 932a68e0

convert: enable --fuse-qkv for muse-glimmer (#29203)

dfriehs committed 29 days ago

Verified 62668d6b

CUDA: tune FA for Gemma 4 on Ampere or newer (#29152)

JohannesGaessler committed 29 days ago

Verified ce8caa6e

metal : support arbitrary hc in dsv4_hc_pre (#29169)

ggerganov committed 29 days ago

Verified a894dae9

common/peg : handle invalid utf-8 sequences in the AST (#29161)

aldehir committed 30 days ago

Verified 3d82ef62

CUDA: enable sparse fa for qwen4 (#28770)

am17an committed 30 days ago

Verified 3cf03257

ui: Fix mobile breakpoint + content overflow issues (#29108)

allozaur committed 30 days ago

Verified b23efaa2

fix(mamba) : make time-step projection input contiguous (#28832)

abetlen committed 30 days ago

Verified 42609036

metal: add F16 input to the FWHT (#29094)

bri-prism committed 30 days ago

Verified 9a9f939b

chat : fix gemma4 required tool grammar (#29115)

aldehir committed 30 days ago

Verified f072b103

Older