ggerganov/llama.cpp

Pull Requests Commits

vulkan: optimizations for direct convolution (#14933)

jeffbolznv committed 1 year ago

Verified a9f7541e

CUDA: fix MMQ nwarps for AMD with warp_size==32 (#15014)

JohannesGaessler committed 1 year ago

Verified 9c35706b

vendor : update vendored copy of google/minja (#15011)

l-austenfeld committed 1 year ago

Verified c76b420e

model : add hunyuan dense (#14878)

stevenkuang-tencent committed 1 year ago

Verified 0f5ccd6f

opencl: add f16 for `add`, `sub`, `mul`, `div` (#14984)

lhez committed 1 year ago

Verified 1c872f71

ggml : Q2k interleaving implementation - x86/x64 SIMD (#14373)

Srihari-mcw committed 1 year ago

Verified baad9488

graph : fix equal_seq() check (#14986)

ggerganov committed 1 year ago

Verified ba42794c

docker : add cann build pipline (#14591)

diannaojiang committed 1 year ago

Verified 2860d479

compare-commits.sh: support both llama-bench and test-backend-ops (#14392)

yeahdongcn committed 1 year ago

Verified 484b2091

quantize : skip tensor override when in fallback mode (#14995)

EAddario committed 1 year ago

Verified daf2dd78

llama : add simple option to enable CPU for MoE weights (--cpu-moe) (#14992)

slaren committed 1 year ago

Verified a06ed5fe

Fix params bug in diffusion example (#14993)

am17an committed 1 year ago

Verified 78452405

llama : allow other bufts when overriding to CPU, add --no-repack option (#14990)

slaren committed 1 year ago

Verified d6818d06

Vulkan: Fix minor debug mode issues (#14899)

0cc4m committed 1 year ago

Verified e08a9882

mtmd : support MiniCPM-V 4.0 (#14983)

tc-mb committed 1 year ago

Verified 952a47f4

MODEL_TENSOR.SSM_DT_NORM has defined twice (#14991)

csabakecskemeti committed 1 year ago

Verified 36e5fe7b

server : implement universal assisted decoding (#12635)

g2mt committed 1 year ago

Verified 94933c8c

llama : merge build_moe_ffn_from_probs function into build_moe_ffn (#14968)

wdl339 committed 1 year ago

Verified c1dacaa9

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

lukasstraub2 committed 1 year ago

Verified a9f77a8b

Add LLaDA 8b Diffusion model (#14771)

am17an committed 1 year ago

Verified 8a4a8562

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

hipudding committed 1 year ago

Verified 11490b36

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

compilade committed 1 year ago

Verified 66625a59

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

lhez committed 1 year ago

Verified 6e672545

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

EAddario committed 1 year ago

Verified e9192bec

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

danbev committed 1 year ago

Verified 41e78c56

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

IMbackK committed 1 year ago

Verified ad4a7001

ggerganov committed 1 year ago

e32a4ec6

cmake : Fix BLAS link interface (ggml/1316)

dg0yt committed 1 year ago

e228de94

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

dg0yt committed 1 year ago

73a8e5ca

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

JohannesGaessler committed 1 year ago

Verified 92b8810e

Newer Older