ggerganov/llama.cpp

Pull Requests Commits

llama : build the kv_rows lists from per-sequence cell bitmaps

am17an committed 4 days ago

006fc936

cont : optimize `get_seq_cells`

ggerganov committed 4 days ago

84323e7e

llama : use llm_attn_mask and kv_rows for unified attention

am17an committed 6 days ago

a2f3d691

gemma3n, gemma4 : create token_embd before the tied output (#31394)

m4zar committed 6 days ago

Verified 344846b4

kv-cache : keep rotation constants on the GPU (#31073)

anton-iermolenko committed 6 days ago

Verified fb599727

memory: refuse rollbacks past the snapshots of the last ubatch (#31368)

ServeurpersoCom committed 6 days ago

Verified ab3768cd

opencl: add initial q2_0 support (#31340)

shaofeiqi committed 6 days ago

Verified e0f142d5

common : end reasoning budget forcing on an end tag from the prompt (#31374)

tdakhran committed 6 days ago

Verified 3c809f44

sycl: FLASH_ATTN_EXT failures (#30761)

mctylr-gh committed 6 days ago

Verified 36a0e2ff

ggml : add batched tensor copy API (#30881)

ggerganov committed 6 days ago

Verified 418f905a

ROCm/HIP: add RDNA4_M family for gfx1170/gfx1171 (#31373)

slojosic-amd committed 7 days ago

Verified eb4e244b

Quant: OCP FP8 E4M3 support (#27512)

ORippler committed 7 days ago

Verified 68d20b88

skills: specify rule about ggml_build_forward_expand (#31375)

ngxson committed 7 days ago

Verified 0893ec98

model: mistral le chonk (#31376)

ngxson committed 7 days ago

Verified 78feec06

llama : reuse the tensor staging buffer and skip the direct-io bounce buffer (#31321)

am17an committed 7 days ago

Verified 0ac1a7fd

server : restore the spec checkpoint for drafts longer than n_rs_seq (#31350)

7konigsberg committed 7 days ago

Verified 7df8b876

ui: keep chats at the bottom until the reader scrolls up (#31311)

allozaur committed 7 days ago

Verified 8ab29fae

ui: show submit button when the chat input has content (#31324)

allozaur committed 7 days ago

Verified f5d4d754

ui: reveal the collapsed sidebar from its logo and toggle it there (#31307)

allozaur committed 7 days ago

Verified 3ed915e3

ui: add MCP group checkboxes in tools settings and fix indeterminate checkbox colors (#31303)

allozaur committed 7 days ago

Verified f8fcef25

ui: request persistent storage on startup (#31299)

allozaur committed 7 days ago

Verified b7cde589

doc: update ops for opencl (#31347)

lhez committed 7 days ago

Verified 3dd0f60b

HIP/CUDA: enable sparse attn on CDNA (#31259)

IMbackK committed 7 days ago

Verified 78febbad

model : Support Limite 1B - Violetto (#29433)

giuseppe committed 7 days ago

Verified abb76ab3

common: select A100 cores for SpaceMiT default threads (#30284)

inuEbisu committed 7 days ago

Verified 41be0bdc

opencl: initial support for iq2 (#31267)

lhez committed 7 days ago

Verified 22183e51

metal : faster MXFP4 mat-vec, full threadgroups for unary/binary/GLU (#31075)

am17an committed 7 days ago

Verified 6d194531

tests : remove duplicate FA test cases added in #31125 (#31357)

Torchit1 committed 7 days ago

Verified fade717e

ggml-metal : fix unary kernels compile on older Metal toolchains (#31360)

ggerganov committed 7 days ago

Verified c9daee14

hexagon: support for UPSCALE op (#31331)

max-krasnyansky committed 7 days ago

Verified 8a199ad3

Older