ggerganov/llama.cpp

Pull Requests Commits

llama : disable tensor parallelism for mellum amax src1

ggerganov committed 4 days ago

981f0a67

vendor : update cpp-httplib to 0.62.0 (#30785)

cabelo committed 4 days ago

Verified 7951f601

moe-cache : replace MoE LRU cache with a 50/50 SLRU cache (#30637)

eapache committed 4 days ago

Verified 33d6135c

metal : fix moe_reduce fusion race when dst aliases its inputs (#30795)

ggerganov committed 4 days ago

Verified 098ce526

cann : refactor `#ifdef` blocks to avoid unreachable code after `return` (#25273)

rauletorresc committed 4 days ago

Verified f3fb2fcd

tests: add big-endian conversion tests (#30705)

taronaeo committed 4 days ago

Verified fda00f18

CUDA: fix warning for nonexistent extra semicolon (#30774)

JohannesGaessler committed 4 days ago

Verified 4866db35

ggml-cpu: vectorize ggml_v_silu and ggml_v_expf (#30791)

taronaeo committed 4 days ago

Verified 42b70032

cuda: cap MMQ Stream-K blocks at available work (#30553)

joelagnel committed 4 days ago

Verified 4c1004c4

qwen3 : bidirectional attention and dense_2 output projection (#30607)

fzowl committed 4 days ago

Verified 412a363c

ui: Fix markdown tables and lists being cut off on phones (#30687)

allozaur committed 4 days ago

Verified 2b57148b

tools : reject truncated GGUF tensor data in split and export tools (#30497)

ppAlexa committed 4 days ago

Verified 8f97d82d

tests : update Metal fusion baseline for beam (#30764)

ggerganov committed 4 days ago

Verified 27af5816

pi : turn .pi/gg into a system prompt addendum (#30762)

ggerganov committed 4 days ago

Verified 7db0d3d7

CUDA: fix shared-memory WAR hazard in flash_attn_ext_vec (#30573)

invaliddayta committed 4 days ago

Verified 99db719b

hexagon: FLASH_ATTN_EXT improvements for longer context and SWA models (#30731)

max-krasnyansky committed 4 days ago

Verified 31288c9c

hexagon: allow multistream FA (#30218)

jhen0409 committed 4 days ago

Verified fe8841a7

cuda : fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain (#30602)

am17an committed 5 days ago

Verified ff5a2337

ggml: support non-contiguous SUM (CPU, CUDA, Vulkan, WebGPU) (#28851)

rssndr committed 5 days ago

Verified 0878f7c3

ggml-cuda: fattn do not check oob if k_VKQ_max = (k_VKQ_max - nbatch_fa) (#30572)

thelittlefireman committed 5 days ago

Verified 5aeac37a

Restore ESIMD kernels on FFN path (#30239)

malsbat committed 5 days ago

Verified 547208dc

hexagon: add support for hadamard MUL_MAT and GET_ROW for flat and tiled Q1_0 and Q2_0 (#30627)

tboinovski1 committed 5 days ago

Verified 848bbda9

hexagon: add support for Q1_0 and Q2_0 (#30122)

tboinovski1 committed 6 days ago

Verified 797de63b

ggml-cuda: HIP: use __builtin_amdgcn_perm for q1_0 unpacking (#29927)

thelittlefireman committed 6 days ago

Verified 9d1878a4

ui : Discover & Download Models from HF Hub (#29584)

allozaur committed 6 days ago

Verified c0fd7862

CUDA: faster MMVQ for low-bit types (#30517)

am17an committed 6 days ago

Verified 2744e48d

tests: give the random model generator a per-tensor filler table (#30674)

ServeurpersoCom committed 6 days ago

Verified 10b8b755

cuda : fix -Wtautological-compare in fattn vec dispatch (#30519)

angt committed 6 days ago

Verified 9d1e0361

cuda : add grouped experts top-k fusion (#29181)

CISC committed 6 days ago

Verified f2f667b0

model : add Kolibri-1 (Aleph-Alpha) support (#30673)

riteme committed 6 days ago

Verified 62318cd5

Older