ggml-org/llama.cpp

Pull Requests Commits

kv-cache : SWA checkpoints store only non-masked cells

ggerganov committed 13 hours ago

Verified faed251c

llama: limit max outputs of `llama_context` (#23861)

am17an committed 15 hours ago

Verified de6f727a

metal: template GLU kernels to support f16/f32 (#23882)

shrivasshankar committed 17 hours ago

Verified 95b8b8ec

vulkan: don't hold the device mutex while compiling pipelines (#23641)

jeffbolznv committed 18 hours ago

Verified 55ac0909

vulkan: reduce host memory lock contention (#23376)

winstonma committed 18 hours ago

Verified bef69f13

vocab: add normalizer.lowercase support to WPM (#23899)

o7si committed 18 hours ago

Verified 5aba5364

TP: quantized KV cache support (#23792)

JohannesGaessler committed 19 hours ago

Verified 8e6fff84

security : disable private disclosures (#23963)

ggerganov committed 19 hours ago

Verified 02a57017

model: Add EXAONE 4.5 implementations (#21733)

nuxlear committed 20 hours ago

Verified 48b88c3b

vulkan: Block-load Q3_K/Q6_K block data and subtract on 32b ints (#23056)

TheBlueMatt committed 20 hours ago

Verified 19620004

vulkan: Removed unused functions (#23175)

winstonma committed 20 hours ago

Verified f8c0a19d

common : support manually triggering the reasoning budget end sequence (#23949)

aldehir committed 20 hours ago

Verified 5254a799

ci : add missing Linux label to cpu-x64-high-perf runner (#23958)

ggerganov committed 22 hours ago

Verified e22b0de6

[SYCL] Support Q4_1, Q5_0, Q5_1 in Flash-attention (#23812)

arthw committed 23 hours ago

Verified a5114249

[SYCL] Add more types in GET_ROWS OP (#23710)

arthw committed 23 hours ago

Verified 41625226

sycl : Optimize Q3_K mul_mat by reorder (#23725)

arthw committed 23 hours ago

Verified 44e211ce

ci: remove redundant or duplicate jobs (#23927)

netrunnereve committed 1 day ago

Verified af6528e6

server : handle If-None-Match weak ETags (#23916)

EZForever committed 1 day ago

Verified 6f165c1c

ci : limit trigger paths for the CPU workflow (#23938)

ggerganov committed 1 day ago

Verified 399739d5

vocab : add tokenizer support for jina-embeddings-v2-base-zh (#18756)

o7si committed 1 day ago

Verified d4c8e2c2

ui: fix ETag truncation with MSVC compiler (#23917)

EZForever committed 1 day ago

Verified 3292da09

docs : update ZenDNN docs for Q8 support (#23791)

truecoder34 committed 1 day ago

Verified e6123e20

llama: only use one iGPU device by default (#23897)

0cc4m committed 1 day ago

Verified 22cadc19

webui: add custom CSS injection via config (#23904)

ServeurpersoCom committed 2 days ago

Verified d749821d

Support `-fa auto` in llama-bench (#23714)

gaugarg-nv committed 2 days ago

Verified aa46bda8

opencl: support bf16 by converting to f16 (#23839)

lhez committed 2 days ago

Verified d6588daa

ui: exclude generated build dirs from prettier and eslint so lint errors stop being masked (#23910)

ServeurpersoCom committed 2 days ago

Verified d38d50e7

TP: fix granularity for Qwen 3.5/3.6 + 3 GPUs (#23843)

JohannesGaessler committed 2 days ago

Verified 8b0e0db6

metal : restore im2col implementation for large kernels (#23901)

ggerganov committed 2 days ago

Verified 2d9b7c8e

test: (test-llama-archs) log the config name first (#23885)

ngxson committed 2 days ago

Verified e674b127

Older