ggml-org/llama.cpp

Pull Requests Commits

iq1_s: slightly faster dot product

Iwan Kawrakow committed 1 year ago

5c977221

Iwan Kawrakow committed 1 year ago

f604a179

iq1_s: Metal works, but quite slow

Iwan Kawrakow committed 1 year ago

425c6bbb

iq1_s: Metal basics

Iwan Kawrakow committed 1 year ago

020b548e

iq1_s: use IQ2_XXS for attn_output

Iwan Kawrakow committed 1 year ago

4be44b7c

iq1_s: better grid

Iwan Kawrakow committed 1 year ago

307c5f61

iq1_s: ARM_NEON dot product. Works, but not very fast

Iwan Kawrakow committed 1 year ago

77301492

iq1_s: AVX2 finally works

Iwan Kawrakow committed 1 year ago

2ffb05ac

Fix after merge with latest master

Iwan Kawrakow committed 1 year ago

67e7c423

Fix shadow warnings

Iwan Kawrakow committed 1 year ago

dc0b14be

Iwan Kawrakow committed 1 year ago

5574533a

iq1_s: WIP AVX2 dot product - something is not right

Iwan Kawrakow committed 1 year ago

592b3b26

iq1_s: scalar CPU dot product

Iwan Kawrakow committed 1 year ago

d94139bf

iq1_s: CUDA is working

Iwan Kawrakow committed 1 year ago

a9d48e97

iq1_s: WIP basics

Iwan Kawrakow committed 1 year ago

80cd5bae

bert : add tests + fix quantization (#5475)

ggerganov committed 1 year ago

Verified 49cc1f7d

tests : disable moe test (#5473)

ggerganov committed 1 year ago

Verified 99b8b43d

ggml-quants : fix compiler warnings (shadow variable) (#5472)

ikawrakow committed 1 year ago

Verified 895407f3

llama : fix quantization when tensors are missing (#5423)

ggerganov committed 1 year ago

Verified 099afc62

swift : package no longer use ggml dependency (#5465)

ggerganov committed 1 year ago

Verified df334a11

py : fix persimmon `n_rot` conversion (#5460)

lx200916 committed 1 year ago

Verified dbd8828e

ggml-sycl: Replace 3d ops with macro (#5458)

abhilash1910 committed 1 year ago

Verified 43fe07c1

llava : remove prog parameter from ArgumentParser (#5457)

danbev committed 1 year ago

Verified 4a46d2b7

sync : ggml (#5452)

ggerganov committed 1 year ago

Verified 3b169441

CUDA: mul_mat_vec_q tiling, refactor mul mat logic (#5434)

JohannesGaessler committed 1 year ago

Verified 3bdc4cd0

Add support for BERT embedding models (#5423)

iamlemec committed 1 year ago

Verified 2891c8aa

flake.lock: Update

github-actions[bot] committed 1 year ago

97a33650

vulkan: only use M-sized matmul on Apple GPUs (#5412)

slp committed 1 year ago

Verified c88c74f9

common : use enums for sampler types (#5418)

z80maniac committed 1 year ago

Verified a803333a

server : allow to specify tokens as strings in logit_bias (#5003)

z80maniac committed 1 year ago

Verified 68478014

Older