ggerganov/llama.cpp

Pull Requests Commits

imatrix : fix 3d tensor counts

compilade committed 35 days ago

91e67b85

Merge branch 'master' into compilade/imatrix-saner-3d

compilade committed 35 days ago

05beb070

mtmd : support MiniCPM-V 4.0 (#14983)

tc-mb committed 35 days ago

Verified 952a47f4

imatrix : fix 3d activations when model tensor is 2d

compilade committed 35 days ago

d4f36e5e

MODEL_TENSOR.SSM_DT_NORM has defined twice (#14991)

csabakecskemeti committed 35 days ago

Verified 36e5fe7b

server : implement universal assisted decoding (#12635)

g2mt committed 35 days ago

Verified 94933c8c

llama : merge build_moe_ffn_from_probs function into build_moe_ffn (#14968)

wdl339 committed 35 days ago

Verified c1dacaa9

server : add openai-style logit_bias support (#14946)

lukasstraub2 committed 35 days ago

Verified a9f77a8b

Add LLaDA 8b Diffusion model (#14771)

am17an committed 35 days ago

Verified 8a4a8562

CANN: Improve loading efficiency after converting weights to NZ format. (#14985)

hipudding committed 35 days ago

Verified 11490b36

graph : reduce splits for recurrent and hybrid models (#14825)

compilade committed 36 days ago

Verified 66625a59

opencl: add `mul_mat_f32_f32_l4_lm` and `mul_mat_f16_f32_l4_lm` (#14809)

lhez committed 36 days ago

Verified 6e672545

quantize : fix using combined imatrix GGUFs (multiple datasets) (#14973)

EAddario committed 36 days ago

Verified e9192bec

server : add support for `embd_normalize` parameter (#14964)

danbev committed 36 days ago

Verified 41e78c56

HIP: enable mfma mmq on gfx908 and gfx90a for select datatypes and shapes (#14949)

IMbackK committed 36 days ago

Verified ad4a7001

ggerganov committed 36 days ago

e32a4ec6

cmake : Fix BLAS link interface (ggml/1316)

dg0yt committed 36 days ago

e228de94

vulkan : fix 32-bit builds (ggml/1313)

dg0yt committed 36 days ago

73a8e5ca

CUDA: skip masked KV slices for all FA kernels (#14924)

JohannesGaessler committed 36 days ago

Verified 92b8810e

tests : update for LLAMA_SET_ROWS=1 (#14961)

ggerganov committed 36 days ago

Verified 00131d6e

graph : fix stack-use-after-return (#14960)

ggerganov committed 36 days ago

Verified 1e15bfd4

embeddings: fix extraction of CLS pooling results (#14927)

iamlemec committed 37 days ago

Verified a118d802

CANN: update ops docs (#14935)

bachelor-dou committed 37 days ago

Verified 61550f82

HIP: remove the use of __HIP_PLATFORM_AMD__, explicitly support only AMD targets (#14945)

IMbackK committed 37 days ago

Verified aa79524c

HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (#14930)

IMbackK committed 37 days ago

Verified b77d1117

HIP: Ignore unsupported unroll transformation in fattn-vec (#14931)

IMbackK committed 37 days ago

Verified c7aa1364

common : avoid logging partial messages (which can contain broken UTF-8 sequences) (#14937)

kallewoof committed 37 days ago

Verified 1a67fcc3

CANN: Add ggml_set_rows (#14943)

hipudding committed 37 days ago

Verified 204f2cf1

cuda : add softcap fusion (#14907)

CISC committed 37 days ago

Verified 138b288b

server-bench: make seed choice configurable (#14929)

JohannesGaessler committed 38 days ago

Verified bbd0f917

Older