ggml-org/llama.cpp

Pull Requests Commits

server: implement disk cache

ngxson committed 50 minutes ago

a9af0e11

use sha256 hash

ngxson committed 2 hours ago

5e37a91c

Merge branch 'master' into xsn/server_slot_sched

ngxson committed 3 hours ago

6b977c88

hexagon: add TIMESTEP_EMBEDDING HVX kernel (#32228)

pdhinaka committed 3 hours ago

Verified dca216ad

memory : fix non-consecutive position warning for M-RoPE (#32287)

nikwen committed 3 hours ago

Verified f8cc5774

llamafile: introduce more Q0 quant support for s390x (#32203)

taronaeo committed 3 hours ago

Verified acf5d545

llama : allocate the tensor staging buffer from the CPU buffer type (#31438)

ykhrustalev committed 3 hours ago

Verified 3e41ae69

vulkan: Explicitly include SPIRV-Headers (#31416)

richiejp committed 3 hours ago

Verified 3bc612fe

vulkan: optimize prompt processing Flash Attention for RDNA3 and RDNA4 (#31323)

0cc4m committed 3 hours ago

Verified d8f30a5b

cuda : keep columns in registers for cooperative softmax (#28982)

sirohikartik committed 3 hours ago

Verified e9641756

vulkan: optimize matrix multiplication for narrow N (#31990)

jeffbolznv committed 3 hours ago

Verified d20426bb

vulkan: use GPU copies on UMA with uncached memory on Intel (#32024)

0cc4m committed 3 hours ago

Verified 19f31e61

sycl : optimize mtmd image processing (#32109)

NickM-27 committed 3 hours ago

Verified 07037ba1

chat : allow undeclared tool calls in LFM2 tool-call grammar (#32164)

tdakhran committed 3 hours ago

Verified 0cc522e4

docs : add OpenVINO backend to ops matrix (#32259)

tokinasin committed 3 hours ago

Verified f08bca8a

chat : allow tool calls inside reasoning blocks for Ministral-3 (#32271)

kukirin-ui committed 3 hours ago

Verified 3c311796

common : fix infinite recursion in debug build for peg parser dump (#32282)

alexzk1 committed 3 hours ago

Verified 8b4ac249

opencl: fixes for q4_k mv oob access, f16 swiglu_oai, cpy with strided dst, batched group_norm and bf16 sigmoid (#32251)

lhez committed 3 hours ago

Verified 1ae047c9

chat : allow Gemma4 tool call during thought (#32232)

cristianadam committed 3 hours ago

Verified 1c2adc75

Fix ctz64 on 32bit (#32273)

miyanyan committed 8 hours ago

Verified a956a852

hexagon: fix fa no-mask tail (#32243)

jhen0409 committed 8 hours ago

Verified c6c0ec6d

qwen2vl : apply FFN weight scales (#32241)

am17an committed 8 hours ago

Verified d0764c5e

convert : register Granite4VisionForConditionalGeneration for text (#32237)

mrutkows committed 11 hours ago

Verified 3549c274

ggml-backend : start a new split at an incompatible weight in a split without inputs (#31775)

am17an committed 12 hours ago

Verified c56be864

convert : pplx-decider must match qwen3.5 model_type (#32156)

jjbuck committed 12 hours ago

Verified a9a5d216

mtmd: fix video pixel aspect ratio and video rotation (#32267)

ngxson committed 12 hours ago

Verified 9d925e7c

mtmd : allow custom video decoder (#32188)

jhen0409 committed 13 hours ago

Verified e3483e40

jinja : reduce template rendering overhead (#32023)

Neroued committed 13 hours ago

Verified 4f2187e7

mtmd : avoid padding-only Whisper audio chunks (#30403)

hera2019 committed 13 hours ago

Verified 0cce7580

spec + model: refactor DFlash based speculative decoding methods (#32016)

ruixiang63 committed 17 hours ago

Verified 32bdc3a8

Older