ggml-org/llama.cpp

Pull Requests Commits

vulkan: fix mul_mat_vec_iq1_s formatting

0cc4m committed 15 hours ago

a5f41199

common : refactor common_sampler + grammar logic changes (#17937)

ggerganov committed 18 hours ago

Verified 254098a2

vulkan: Fix data race/hang in scalar/cm1 flash attention (#17887)

jeffbolznv committed 19 hours ago

Verified 3238b140

vulkan: improve mul_mat_vec_iq1_s speed (#17874)

lovedheart committed 19 hours ago

Verified 47226716

vulkan: faster q6_k matmul (#17813)

netrunnereve committed 19 hours ago

Verified d15d177f

model-conversion : cast logits to float32 (#18009)

ggerganov committed 20 hours ago

Verified 77ad8542

models : fix YaRN regression + consolidate logic (#18006)

ggerganov committed 20 hours ago

Verified 609a2d02

ggml : arm repack fix build

ggerganov committed 20 hours ago

a63cbafb

ggerganov committed 20 hours ago

0e592249

ggml : arm repack fix build (whisper/0)

ggerganov committed 20 hours ago

71fdcf06

cmake : set `CMAKE_RUNTIME_OUTPUT_DIRECTORY` for non standalone build (ggml/1394)

HerrCai0907 committed 20 hours ago

615655aa

scripts: add script to compare logprobs of llama.cpp against other frameworks (#17947)

ngxson committed 1 day ago

Verified c00ff929

server-models.cpp: add missing <filesystem> (#18000)

barracuda156 committed 1 day ago

Verified 4ed2bae5

llama_context: synchronize before reallocating output buffer (#17974)

jeffbolznv committed 1 day ago

Verified 5266379b

arg: fix common_params_parse not accepting negated arg (#17991)

ngxson committed 1 day ago

Verified 4d5ae24c

cmake: correct scope - link ws2_32 for MinGW/w64devkit builds in cpp-httplib (#17972)

gustrd committed 1 day ago

Verified 66ba5125

vulkan: support get_rows for i32 (#17941)

jeffbolznv committed 1 day ago

Verified 36255a22

vulkan: support GGML_OP_DIAG (#17893)

jeffbolznv committed 1 day ago

Verified 3229a23f

vulkan: Multi-pass softmax for large number of cols (#17892)

jeffbolznv committed 1 day ago

Verified 303f8615

speculative-simple : free batch on exit (#17985)

ggerganov committed 1 day ago

Verified 3c6391e7

common : skip model validation when --completion-bash is requested (#17975)

CISC committed 1 day ago

Verified 8e4d6785

vulkan: Allow non-pow2 n_experts in topk_moe (#17872)

jeffbolznv committed 1 day ago

Verified 07a10c10

add llama-completion to completion-bash executables (#17976)

CISC committed 1 day ago

Verified 2bc94e79

model-conversion : use CONVERTED_MODEL value for converted model [no ci] (#17984)

danbev committed 1 day ago

Verified fd1085ff

common: support negated args (#17919)

ngxson committed 2 days ago

Verified 380b4c98

clip: move model cgraphs into their own files (#17965)

ngxson committed 2 days ago

Verified e39a2ce6

ci : change the cann version and the container pull method (#17953)

xuedinge233 committed 2 days ago

Verified a8c7f33d

docker : include legacy llama-completion binary (#17964)

CISC committed 2 days ago

Verified b7f5f46e

CUDA: fix overflow in MMA kernel without stream-k (#17939)

JohannesGaessler committed 2 days ago

Verified 48221143

models : fix the attn_factor for mistral3 graphs + improve consistency (#17945)

ggerganov committed 2 days ago

Verified 7bed317f

Older