ggml-org/llama.cpp

Pull Requests Commits

rpc : use std::unique_ptr for the message_queue

ggerganov committed 48 days ago

Verified 091d98e2

rpc : implement event and async backend APIs

rgerganov committed 49 days ago

df27d80a

server : fix router child env in containerized environments (#18562)

v-sayapin committed 49 days ago

Verified da143b99

vulkan: fix topk_moe_sigmoid_norm_bias failures in GLM-4.6 (#18582)

jeffbolznv committed 49 days ago

Verified f1768d8f

models : fix backend assignment for Granite/Nemotron graphs (#18599)

ggerganov committed 49 days ago

Verified 2da64a2f

vulkan: handle quantize_q8_1 overflowing the max workgroup count (#18515)

jeffbolznv committed 49 days ago

Verified b37124d2

llama : refactor rope_freq_base/scale_swa conversion and init (#18553)

CISC committed 49 days ago

Verified eadc4184

CANN: add operator fusion support for ADD + RMS_NORM (#17512)

noemotiovon committed 49 days ago

Verified 67e3f6f6

doc: clarify that steps also apply to linux for opencl (#18002)

alosslessdev committed 49 days ago

Verified 92ac1e01

ci : init git lfs in every build for RISC-V (#18590)

alitariq4589 committed 49 days ago

Verified 8e3a7611

sampling : add support for backend sampling (#17004)

danbev committed 50 days ago

Verified d3dce4e0

model : mtmd : make input norm optional in LFM2-VL (#18594)

tdakhran committed 50 days ago

Verified 4974bf53

CUDA: disable cuda graph when using n-cpu-moe (#18593)

am17an committed 50 days ago

Verified 908a9e5a

ggml-cuda: remove unused params in ggml_cuda_graph (#18579)

am17an committed 50 days ago

Verified 5126c41c

common/grammar : replace problematic backtracking regex `[\s\S]*` (#18342)

aldehir committed 50 days ago

Verified cef1d23c

graph : fix graph reuse logic when `n_pos_per_embd > 1` (#18566)

ggerganov committed 50 days ago

Verified c69c7ebc

ggml-cuda: fixes for concurrent streams (#18496)

am17an committed 51 days ago

Verified e57f5233

context : fix reserve token padding to n_seqs (#18536)

ggerganov committed 51 days ago

Verified a554a1ec

CUDA: only allocate FA tmp buffer if needed (#18564)

JohannesGaessler committed 51 days ago

Verified 0f2e42ca

(Bugfix, ggml-cuda) Pool alloc count fix + small size computation type adjustment (#18559)

pl752 committed 51 days ago

Verified 9dba9f53

ggml-hexagon: optimize activation function (#18393)

joeldushouyu committed 51 days ago

Verified bcfc8c3c

vulkan: Optimize GGML_OP_CUMSUM (#18417)

jeffbolznv committed 51 days ago

Verified 18ddaea2

vulkan: Implement mmvq for iq1_s/iq1_m (#18450)

jeffbolznv committed 52 days ago

Verified 706e3f93

model : Maincoder-1B support (#18534)

maincode-prabod committed 52 days ago

Verified 5755e52d

metal : adjust extra size for FA buffer to avoid reallocations (#18545)

ggerganov committed 52 days ago

Verified f38de163

graph : reduce topology branching (#18548)

ggerganov committed 52 days ago

Verified af1e8e1a

vocab : reduce debug logs about non-EOG control tokens (#18541)

ggerganov committed 52 days ago

Verified d84a6a98

rpc : use unordered_map::reserve and emplace (#18513)

struct committed 52 days ago

Verified c6f0e832

cuda : fix copy of large tensors (ggml_nbytes <= INT_MAX assertion) (#18433)

Meet91721 committed 52 days ago

Verified e86f3c22

model : remove modern-bert iswa template (#18529)

CISC committed 52 days ago

Verified 169ee68f

Older