ggml-org/llama.cpp

Pull Requests Commits

context : fix logits size overflow for huge batches

compilade committed 1 year ago

145401c9

context : fix overflow when re-ordering huge outputs

compilade committed 1 year ago

f16a843a

llama : add --n-cpu-moe option (#15077)

slaren committed 1 year ago

Verified ec428b02

imatrix : warn when GGUF imatrix is saved without .gguf suffix (#15076)

compilade committed 1 year ago

Verified 19f68fa5

cmake: Add GGML_BACKEND_DIR option (#15074)

ckastner committed 1 year ago

Verified 41613437

gguf-py : add --chat-template-file to gguf_new_metadata (#15075)

CISC committed 1 year ago

Verified e5bebe52

model: support GLM 4.5 family of models (#14939)

sammcj committed 1 year ago

Verified ef0144c0

quantize : fix confusing error message if ftype is invalid (#15071)

CISC committed 1 year ago

Verified 2721257e

ggml: WebGPU backend host improvements and style fixing (#14978)

reeselevine committed 1 year ago

Verified 587d0118

vulkan: fix build when using glslang that does not support coopmat2 (#15062)

jeffbolznv committed 1 year ago

Verified 5aa1105d

imatrix : use GGUF by default (#14842)

compilade committed 1 year ago

Verified d31192b4

imatrix : fix 3d activation handling for hybrid and recurrent models (#14994)

compilade committed 1 year ago

Verified 0a2f5496

memory : handle kv_unified for hybrid models (#15050)

compilade committed 1 year ago

Verified 11a38111

vocab : JetBrains Mellum pre-tokenizer (#15045)

csabakecskemeti committed 1 year ago

Verified 97366dc6

model : add text-only support for Kimi-VL (and find special tokens in text_config) (#15051)

gabriellarson committed 1 year ago

Verified 83bc2f28

vulkan: Use coopmat2 for conv2d (#14982)

jeffbolznv committed 1 year ago

Verified 6c7a4411

opencl: fix adreno compiler detection logic (#15029)

lhez committed 1 year ago

Verified 5c0eb5ef

CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)

JohannesGaessler committed 1 year ago

Verified 03d46982

cuda: make im2col a little faster (#15025)

leejet committed 1 year ago

Verified 3303c19b

kv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)

danbev committed 1 year ago

Verified 4fdea540

llama : enable LLAMA_SET_ROWS=1 by default (#14959)

ggerganov committed 1 year ago

Verified a4569c41

cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (#15038)

ggerganov committed 1 year ago

Verified 15e92fd3

ci : check that pre-tokenizer hashes are up-to-date (#15032)

CISC committed 1 year ago

Verified 2bf3fbf0

convert : fix Qwen3-Embedding pre-tokenizer hash (#15030)

iamlemec committed 1 year ago

Verified 711d5e6f

chat : fix multiple tool_calls on hermes-2-pro (#14962)

jhen0409 committed 1 year ago

Verified f738989d

vulkan: coopmat2 mul_mat optimizations (#14934)

jeffbolznv committed 1 year ago

Verified 4cb208c9

llama-bench: rename DB table name from test to llama_bench (#15003)

yeahdongcn committed 1 year ago

Verified 3025b621

vulkan: Support ne[3]>1 in noncontig matrix-vector multiply (#15015)

jeffbolznv committed 1 year ago

Verified ec0b1880

model : support Qwen3-Embedding (#15023)

iamlemec committed 1 year ago

Verified 339bd026

server: enable token array inputs for OAI API (#15001)

JohannesGaessler committed 1 year ago

Verified f9062755

Older