ngxson/llama.cpp

Pull Requests Commits

restore to initial min/max tokens value for qwen

ngxson committed 55 days ago

a834e1ce

no resize if dst size == src size

ngxson committed 55 days ago

00ee52e7

qwen: increase min tokens

ngxson committed 55 days ago

29c72688

use simple resize for qwen

ngxson committed 55 days ago

bae84d44

gonna fix them all

ngxson committed 56 days ago

4621d99d

fix llama4 and lfm2

ngxson committed 56 days ago

2c0d9607

get_merge_kernel_size()

ngxson committed 56 days ago

bfd03fb6

fix token calculation

ngxson committed 56 days ago

000d1d9d

ngxson committed 56 days ago

42744178

fix out of bound composite

ngxson committed 56 days ago

2892e0fb

ngxson committed 56 days ago

7bd1a011

better image preproc for qwen

ngxson committed 56 days ago

68b1507a

improve hparams

ngxson committed 56 days ago

c53c5663

implement mix/max pixels

ngxson committed 56 days ago

66d5c43d

ngxson committed 56 days ago

13cd2045

mtmd: refactor preprocessing + support max/min pixels

ngxson committed 56 days ago

c1b189ae

convert : update transformers requirements (#16866)

RodriMora committed 56 days ago

Verified ce18efea

server : bump request URI max length to 32768 (#16862)

chansikpark committed 57 days ago

Verified 16724b5b

server : remove n_past (#16818)

ggerganov committed 57 days ago

Verified b52edd25

cpu: introduce chunking for repack matmuls and enable matmul-id chunking on ARM64 (#16833)

max-krasnyansky committed 57 days ago

Verified 517b7170

common: fix typo in cli help text (#16864)

sbera77 committed 57 days ago

Verified 835e918d

model: add support for qwen3vl series (#16780)

JJJYmmm committed 57 days ago

Verified d261223d

cpu: introduce chunking for flash attention (#16829)

max-krasnyansky committed 57 days ago

Verified dcca0d3a

model: Add support for CogVLM model (#15002)

Tianyue-Zhao committed 57 days ago

Verified bacddc04

cuda : fix argsort with 64k+ rows (#16849)

CISC committed 57 days ago

Verified 229bf686

llama : use std::abs instead of abs (#16853)

kaetemi committed 57 days ago

Verified d7395115

vulkan: Handle argsort with a large number of rows (#16851)

jeffbolznv committed 57 days ago

Verified 052df28b

Hide latency of bias and gate-loading (#16847)

ORippler committed 57 days ago

Verified 8b11deea

vulkan: Fuse rope+set_rows (#16769)

jeffbolznv committed 57 days ago

Verified b9ce9401

llama: fix ASAN error with M-RoPE (#16848)

ngxson committed 58 days ago

Verified 3464bdac

Older