ggml-org/llama.cpp

Pull Requests Commits

ggml: asynchronous scheduler memory copies

JohannesGaessler committed 46 days ago

124d3d94

spec: support spec-draft-p-min in DFlash (#25246)

ruixiang63 committed 47 days ago

Verified 152d337f

cuda: enable topk-moe fusion for 288 experts (#25267)

pwilkin committed 47 days ago

Verified 75a48a90

ui: align persisted config with strict server schema and enable thinking by default (#25242)

ServeurpersoCom committed 48 days ago

Verified 067de937

server + ui: ping silent SSE streams every 1s and kick only after 3s so slow prefill never drops healthy connections (#25241)

ServeurpersoCom committed 48 days ago

Verified b5315e16

ui: Add MCP Servers Opt-In for first time visitors (#25239)

allozaur committed 48 days ago

Verified 94875285

Remove redundant CUDA copies after gated_delta_net. (#23940)

gaugarg-nv committed 48 days ago

Verified 5a460dea

vendor : update cpp-httplib to 0.49.0 (#25218)

cabelo committed 48 days ago

Verified c8ae9a75

llama : add llama_model_ftype_name() (#25134)

angt committed 48 days ago

Verified fdb1db87

opencl: allow loading precompiled binary kernels from library (#23042)

lhez committed 49 days ago

Verified 4fc4ec55

common : use hf primary split as model path (#25194)

angt committed 49 days ago

Verified a6647b1a

hexagon: flash attention rework (optimizations, accuracy improvements, etc) (#25085)

max-krasnyansky committed 49 days ago

Verified 13e67386

CUDA: consistent use of __restrict__ + PDL for FA (#25185)

JohannesGaessler committed 50 days ago

Verified b820cc8e

ggml-cpu: add AVX2 optimization for nvfp4 dot product and use UE4M3 LUT (#23961)

ragz4125 committed 50 days ago

Verified 6dbc1174

ui Prevent tool messages from incorrectly appending to other conversations (#25177)

allozaur committed 50 days ago

Verified 9d88e7ce

ui: Remove PWA navigate fallback to prevent caching API endpoint requests (#25174)

allozaur committed 50 days ago

Verified 7af4279f

opencl: initial q1_0 support (#25160)

lhez committed 50 days ago

Verified fd1a0579

cuda : prevent integer truncation and overflow errors when using KQ mask strides in flash_attn_mask_to_KV_max kernel (#24945)

fairydreaming committed 50 days ago

Verified 0eca4d49

model : register t_layer_inp for qwen3next (#25141)

jschmied committed 50 days ago

Verified 4f31eedb

common,server: handle bracketed IPv6 literals in URL authority (#25140)

ServeurpersoCom committed 50 days ago

Verified 799fcc04

CUDA: fix get_rows_back for tables with more than 65535 rows (grid-y clamp + stride) (#25103)

mattjallo committed 51 days ago

Verified 931eb37f

CUDA: fix Gemma E4B MTP FlashAttention (#25148)

JohannesGaessler committed 51 days ago

Verified e495d1e7

vulkan: roll bk loop in matmul for asahi linux (#24663)

xingjianll committed 51 days ago

Verified f708a5b2

HIP: use hipBLAS for dense prefill on gfx900, keep MMQ for MoE (#24588)

DEV-DUFORD committed 51 days ago

Verified d9df1100

ggml-webgpu: add support for NVFP4 (#25143)

yomaytk committed 51 days ago

Verified 6c5de1cc

Revert "sched : reintroduce less synchronizations during split compute (#20793)" (#25138)

ORippler committed 51 days ago

Verified 86b94708

common : dedup preset and cached model entries in /v1/models (#25131)

angt committed 51 days ago

Verified 6f4f53f2

vulkan: use flops instead of weight tensor size for submission heuristic (#25005)

0cc4m committed 51 days ago

Verified 25a1d63f

DeepSeek V4 (#24162)

am17an committed 52 days ago

Verified 8c146a83

tools/ui: restore Tailwind scanning in ignored worktrees (#24879)

seryogakovalyov committed 52 days ago

Verified 6cb18b2f

Older