ggerganov/llama.cpp

Pull Requests Commits

cuda : alternative q4_q8 kernel

ggerganov committed 2 years ago

a3e6d622

More int mult, less float mult, worse performance

JohannesGaessler committed 2 years ago

e7b9d97b

Performance no longer terrible

JohannesGaessler committed 2 years ago

d882d1c2

JohannesGaessler committed 2 years ago

4b128813

fixup! Store layers in VRAM

JohannesGaessler committed 2 years ago

8a9d7ce6

Store layers in VRAM

JohannesGaessler committed 2 years ago

3ed4588e

Faster than CPU without 80% runtime memcpy

JohannesGaessler committed 2 years ago

d052a0ed

Works but slower than CPU

JohannesGaessler committed 2 years ago

229aa1f5

Interface improvements and `--multiline-input` (previously `--author-mode`) (#1040)

DannyDaemonic committed 2 years ago

Verified 41654efe

readme : add notice about upcoming breaking change

ggerganov committed 2 years ago

Verified 56551bc1

readme : add TOC and Pygmalion instructions (#1359)

AlpinDale committed 2 years ago

Verified fe60904e

llama : fix hparams shadow (#1367)

prusnak committed 2 years ago

Verified 003ba2fb

llama : require first token to be BOS (#1303)

ggerganov committed 2 years ago

Verified f9a63649

convert: add ability to convert safetensors files (#1276)

ubik2 committed 2 years ago

Verified 95078cc5

Documented CUDA reproducibility, added warning (#1346)

JohannesGaessler committed 2 years ago

Verified 1f48b0ab

CI: add Windows CLBlast and OpenBLAS builds (#1277)

SlyEcho committed 2 years ago

Verified e1295513

ggml : Allow usage of CLBlast alongside Accelerate.framework (#1336)

swittk committed 2 years ago

Verified 1b0fd454

Remove default arguments from sampling functions (#1343)

j-f1 committed 2 years ago

Verified 39240885

makefile: automatic Arch Linux detection (#1332)

daniandtheweb committed 2 years ago

Verified 173d0e64

ci : add cublas to windows release (#1271)

Green-Sky committed 2 years ago

Verified a3b85b28

readme: add missing info (#1324)

prusnak committed 2 years ago

Verified 921dcee0

Fix for OpenCL / clbast builds on macOS. (#1329)

IonoclastBrigham committed 2 years ago

Verified 2d13786e

Convert.py @staticmethod (#1327)

blecaillon committed 2 years ago

Verified a90e96b2

quantize: make output filename optional, default to ggml-model-<ftype>.bin (#1301)

slaren committed 2 years ago

Verified 94c5652f

Wrap exceptions in std::exception to verbose output on exception. (#1316)

ivanstepanovftw committed 2 years ago

Verified 34d9f22f

convert: support DT_BF16 tensors (#1309)

ivanstepanovftw committed 2 years ago

Verified d3e8093e

readme : add OpenBuddy link (#1321)

44670 committed 2 years ago

Verified 360cfe5b

main : add --in-suffix option (#1318)

44670 committed 2 years ago

Verified 2edbdb0f

ggml : change immintrin.h to intrin.h for compatibility (#1307)

iRonJ committed 2 years ago

Verified 20fbf2a2

Only escape prompts when used with `-e` (#1311)

DannyDaemonic committed 2 years ago

Verified db108087

Older