ngxson/llama.cpp

Pull Requests Commits

chat : fix translategemma crash on common_chat_format_example

ngxson committed 113 days ago

96bd193e

CUDA: add gqa_ratio 4 for GLM 4.7 flash (#18953)

am17an committed 113 days ago

Verified b70d2510

opencl: add TRI op support (#18979)

shaofeiqi committed 113 days ago

Verified 5516b9c1

ggml-zdnn : mark zDNN buffers as non-host (#18967)

AlekseiNikiforovIBM committed 113 days ago

Verified 94242a62

ci : update GitHub Actions versions [no ci] (#18935)

pgoslatara committed 113 days ago

Verified 6b99a223

convert : add Devstral-2 (Ministral3ForCausalLM) arch (#18972)

emsi committed 113 days ago

Verified 77078e80

jinja: support none|string (#18995)

pwilkin committed 113 days ago

Verified c301172f

fix: Use `tabular-nums` for chat message statistics (#18915)

nathanlesage committed 113 days ago

Verified 3802d3c7

llama : clarify nemotron-h.cpp comment about RoPE [no ci] (#18997)

danbev committed 113 days ago

Verified 9da3dcd7

vulkan: Remove transfer_ctx, do everything in compute_ctx. (#18945)

jeffbolznv committed 113 days ago

Verified bd544c94

common : improve error message when HTTPS is missing but required (#18987)

angt committed 113 days ago

Verified 14be5a39

server: /v1/responses (partial) (#18486)

openingnow committed 113 days ago

Verified fbbf3ad1

vulkan: support flash attention GQA/split_k with small batches (#18938)

jeffbolznv committed 113 days ago

Verified 33f890e5

Revert "vulkan: force full subgroups for flash attention to fix intel subgroup crash (#17356)" (#18831)

rillomas committed 113 days ago

Verified 067b8d7a

vulkan: Use mul_mat_vec_id for small values of n (#18918)

jeffbolznv committed 113 days ago

Verified 50b7f076

memory : add llama_memory_hybrid_iswa (#18601)

tdakhran committed 114 days ago

Verified ad8d85bd

Fix GLM 4.7 Lite MoE gating func (#18980)

pwilkin committed 114 days ago

Verified 12a4a47e

gguf: display strerrno when cant load a model (#18884)

teto committed 114 days ago

Verified 37c35f0e

CUDA: Fix builds for older CCCL versions by ifdefing strided_iterator (#18964)

ORippler committed 114 days ago

Verified 5bd341c9

common, server : use the same User-Agent by default (#18957)

angt committed 114 days ago

Verified 1c7cf94b

cli : fix reasoning responses in CLI (#18961)

ngxson committed 114 days ago

Verified 2c1f1996

CUDA: Replace init_offsets kernel with iterators in cub-based argsort (#18930)

ORippler committed 115 days ago

Verified d1e35564

ggml : cleanup path_str() (#18928)

angt committed 115 days ago

Verified 08f3f4a8

metal : enable FA for MLA heads (#18950)

ggerganov committed 115 days ago

Verified 27119190

convert : use n_groups instead of hardcoded values in reshape (#18929)

danbev committed 115 days ago

Verified 7dee9ff5

server : refactor oai_parser_opt, move it to server_chat_params (#18937)

ngxson committed 115 days ago

Verified 6df686be

convert : support Glm4MoeLite (#18936)

ddh0 committed 115 days ago

Verified 1706a6d7

jinja : fix undefined keys and attributes and int/float as bool (#18924)

CISC committed 115 days ago

Verified 959ecf7f

ci : run test-jinja -py on high perf [no ci] (#18916)

CISC committed 115 days ago

Verified 4037093c

server: fix memory reservations in populate_token_probs (#18787)

l-austenfeld committed 115 days ago

Verified 18361c57

Older