ngxson/llama.cpp

Pull Requests Commits

ngxson committed 33 days ago

768416a4

rename to compare-logprobs

ngxson committed 34 days ago

2f79a106

ngxson committed 35 days ago

7ae43cb9

ngxson committed 35 days ago

d41c8bdd

use abs for diff

ngxson committed 35 days ago

bff275ea

ngxson committed 35 days ago

0b4efb6b

clarify endpoint

ngxson committed 35 days ago

a92a8f60

ngxson committed 35 days ago

dc18bfcb

accept custom prompt file

ngxson committed 35 days ago

d6b41ff4

scripts: add script to compare logits of llama.cpp against other frameworks

ngxson committed 35 days ago

f9b98810

Add a search field on model selector / improve mobile display (#17765)

ServeurpersoCom committed 35 days ago

Verified a81a5695

SOLVE_TRI extension to more dimensions (#17793)

pwilkin committed 35 days ago

Verified 53ecd4fd

ggml-alloc : fix reuse-parent logic for misaligned sizes (#17884)

ggerganov committed 35 days ago

Verified c6f6e4f9

batch : fix sequence id ownership (#17915)

ggerganov committed 35 days ago

Verified d9f8f606

docs: use port 8080 in Docker examples (#17903)

utsumi-fj committed 35 days ago

Verified e4ae3833

ggml-hexagon: fix `rope` failure at `test-backend-ops` (#17565)

chraac committed 36 days ago

Verified 34ce48d9

ci: fix riscv64-native build (#17916)

CISC committed 36 days ago

Verified 45e350e3

mtmd: some small clean up (#17909)

ngxson committed 36 days ago

Verified c6b2c931

cli: enable jinja by default (#17911)

ngxson committed 36 days ago

Verified 34a6d869

server: add presets (config) when using multiple models (#17859)

ServeurpersoCom committed 36 days ago

Verified f32ca51b

Fix race conditions in threadpool when dealing with dynamic/frequent n_threads changes (#17748)

max-krasnyansky committed 36 days ago

Verified e1f49219

ggml : remove GGML_KQ_MASK_PAD constant (#17910)

ggerganov committed 36 days ago

Verified 4dff236a

cuda : add missing support check for xielu (#17895)

CISC committed 36 days ago

Verified 4df6e859

cli: new CLI experience (#17824)

ngxson committed 36 days ago

Verified 6c213177

model : Qwen3-Next-80B-A3B has 48 layers (#17898)

EZForever committed 36 days ago

Verified b6777218

docs : update opencl ops (#17904)

lhez committed 36 days ago

Verified 2d2e1030

CUDA: fix unpadded strides in MMA FA kernel (#17891)

JohannesGaessler committed 36 days ago

Verified 17f7f4ba

convert: allow using quantized Mistral weight (#17889)

ngxson committed 36 days ago

Verified 9e79b011

fix softmax for iGPU (#17838)

NeoZhangJianyu committed 36 days ago

Verified 2e9eab80

common : add parser for ministral/mistral large 3/devstral 2 (#17713)

aldehir committed 37 days ago

Verified 2fbe3b7b

Older