ggml-org/llama.cpp

Pull Requests Commits

revert unrelated changes

ngxson committed 33 days ago

44d55b2c

server: use port 9931 by default

ngxson committed 33 days ago

2cf5a955

llama: support MoE cache over multiple GPUs (#30112)

am17an committed 33 days ago

Verified c811cb8f

server : preserve context checkpoints across slot save/restore (#26004)

Tough-Respawn committed 33 days ago

Verified 033df86b

vulkan : fix TOP_K for +inf/NaN inputs and k = 1 on negative values (#30107)

gianni-cor committed 33 days ago

Verified ff5888f9

vulkan: extend sparse FA support to coopmat2 (#30003)

jeffbolznv committed 33 days ago

Verified dac30873

vendor : update cpp-httplib to 0.60.1 (#30134)

angt committed 33 days ago

Verified 03aa006a

cuda : support arbitrary striding for unary ops on f16, f32, and bf16 (#29781)

saady789 committed 33 days ago

Verified 08246a28

sycl: FWHT optimizations (#29605)

bri-prism committed 33 days ago

Verified 46baf1f1

hex-mmadd: do not assume aligned read/write when bias-add is fused (#30133)

max-krasnyansky committed 33 days ago

Verified 097f5b53

cuda : Use byte strides for roll to allow non-contiguous ROLL operations (#29547)

bertaye committed 33 days ago

Verified d8880160

CUDA: fix norm family kernels when ne[2]/ne[3] exceed grid dim limits (#28175)

edenfunf committed 33 days ago

Verified fda18666

sycl: fuse the delta-net alpha gate (add + unary + mul) (#29687)

Titaniumtown committed 33 days ago

Verified ff30363a

sycl: stage bulk uploads (model loading) through a pinned ring buffer (#29608)

cwriter committed 33 days ago

Verified 000bee54

model : support classifier_activation for rerankers (#29692)

boshjerns committed 33 days ago

Verified 37ac6345

ggml-cuda: assign four GDN state columns per warp (#30087)

SongXiaoXi committed 33 days ago

Verified 24e41838

sycl: add grouped MoE XMX GEMM (#29245)

cwriter committed 34 days ago

Verified 847f447c

convert : support Qwen3.5 embedding models (#27920)

SamMalayek committed 34 days ago

Verified 75118a3a

sycl: remove duplicate block-size defines from op headers (#29507)

Titaniumtown committed 34 days ago

Verified 9b4ed0ca

hexagon: enable alloc_buffer_n (#30126)

max-krasnyansky committed 34 days ago

Verified 9c2e0e49

hexagon: Q6_K weight dequant speedup (#30121)

kurquhar committed 34 days ago

Verified 06cad0b9

model : add LiquidAI/d1-omni-600M decision model (#30114)

tdakhran committed 34 days ago

Verified a657f7e9

hexagon: support tiled Q4_K and Q6_K GET_ROWS (#30115)

kurquhar committed 34 days ago

Verified aa5e0092

hexagon: improved GELU accuracy (#30104)

kurquhar committed 34 days ago

Verified 70815103

chat: fix jinja parser for TranslateGemma (#30096)

nbud committed 34 days ago

Verified bd4eeaa0

chat : name tool and argument parser rules by index (#30088)

Frost-54 committed 34 days ago

Verified 5de73343

model : add LiquidAI/d1-3B decision model (#30110)

tdakhran committed 34 days ago

Verified 88dcc460

cuda: FWHT kernels for block widths above 512 (#29100)

bri-prism committed 34 days ago

Verified b86d2f07

mtmd: add cohere2 vision support (#30062)

Terrencezzj committed 34 days ago

Verified 50a6c5cf

llama : add a GPU cache for MoE experts kept in host memory (#29887)

am17an committed 34 days ago

Verified d6cf9acb

Older