ggerganov/llama.cpp

Pull Requests Commits

Merge branch 'master' into xsn/mtmd_ds_ocr_tiles

ngxson committed 33 days ago

1f66d544

test-deepseek-ocr: relax v1 single-view tolerance; drop trailing prompt space; make DRY opt-in and n_predict model-specific (#25486)

sfallah committed 33 days ago

Verified 83c7477b

feat: pre-select models in the webui using alias (#25492)

felix314159 committed 33 days ago

Verified 07d93782

ui: use server modalities in non-router mode (#24874)

coder543 committed 33 days ago

Verified 9f623c68

server: remove loading.html (#25500)

ngxson committed 33 days ago

Verified a935fbff

ggerganov committed 33 days ago

0badc06a

ggml : use ggml_vqtbl1q_u8 for 32-bit compat (whisper/0)

ggerganov committed 33 days ago

ac17f8ac

server: improve tools, remove apply_diff (#25498)

ngxson committed 33 days ago

Verified c4ae9a88

cli: fix crash on wrong server base url (#25497)

marcoStocchi committed 33 days ago

Verified 1b9691bc

ui: prevent tooltip from flickering open and closed on hover (#25503)

ServeurpersoCom committed 33 days ago

Verified c7af942e

sync : ggml (#25517)

ggerganov committed 33 days ago

Verified 8f114a9b

ui: export full message tree instead of active path only (#25501)

ServeurpersoCom committed 33 days ago

Verified d46786f2

llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4 (#25370)

fairydreaming committed 33 days ago

Verified 2ed3c1ab

ggml-et: Initial ET backend (#24179)

marty1885 committed 34 days ago

Verified 082b326f

llama-batch: add unit test (#25471)

am17an committed 34 days ago

Verified 961e4b26

opencl: cluster-parallel decode FA for Adreno (#25473)

wanghqc committed 34 days ago

Verified 049326a0

ggml : process data in smaller chunks in CUDA ggml_top_k() and ggml_argsort() to reduce temporary buffers memory usage (#24776)

fairydreaming committed 34 days ago

Verified 07494499

cli: add --output option (#25484)

ngxson committed 34 days ago

Verified 3de7dd4c

hexagon: tiling, tracing and optimizations for unary ops (#25474)

aparmp-quic committed 34 days ago

Verified fb30ba9a

server : move chat-template thinking probe inside the init try/catch (#24093)

palios-taey committed 34 days ago

Verified 82fce65d

ggml : fix conv 2d dw (#25490)

ggerganov committed 34 days ago

Verified 5c3a5868

meta: add hard emphasis on agents not writing descriptions/comments (#25480)

pwilkin committed 34 days ago

Verified c15c5c77

Refactor: Consistently use smart pointers in `test-backend-ops` (#25440)

ORippler committed 34 days ago

Verified f84a5194

mtmd: dsocr-tiles fixes (#25481)

sfallah committed 34 days ago

Verified e5d8b468

Only index by compile times + always multiply/add (#25445)

ORippler committed 34 days ago

Verified 683f0c72

llama-bench : init params.offline (#25476)

angt committed 34 days ago

Verified 259f2e2a

metal : add CONV_2D_DW (depthwise convolution) support (#21565)

Sou-ly committed 34 days ago

Verified 92b187c9

ggml-hip: enable -funsafe-math-optimizations (#24668)

RapidMark committed 34 days ago

Verified ccb0c342

cuda: align snake fusion matcher with the other backends (#25460)

ServeurpersoCom committed 34 days ago

Verified 2021515a

server : respect min-step when splitting prompt batches (#25420)

aldehir committed 34 days ago

Verified 64c8b7db

Older