ngxson/llama.cpp

Pull Requests Commits

remove next_metadata_size

ngxson committed 2 years ago

38d0e1da

ngxson committed 2 years ago

83f944c2

error on 0 tensors

ngxson committed 2 years ago

b7cb3bb7

add dry run option

ngxson committed 2 years ago

8569ba30

ngxson committed 2 years ago

583022c5

clean up arg parse

ngxson committed 2 years ago

843bc30c

split by max size

ngxson committed 2 years ago

12b25548

[SYCL] Fix batched impl for NVidia GPU (#6164)

AidanBeltonS committed 2 years ago

Verified e82f9e2b

Make IQ1_M work for QK_K = 64 (#6327)

ikawrakow committed 2 years ago

Verified cbc83436

common : change --no-penalize-nl to --penalize-nl (#6334)

CISC committed 2 years ago

Verified e562b971

llama2c : open file as binary (#6332)

ggerganov committed 2 years ago

Verified 2ab4f00d

readme : add php api bindings (#6326)

mcharytoniuk committed 2 years ago

Verified 1740d6dd

server: public: use relative routes for static files (#6325)

EZForever committed 2 years ago

Verified 0642b22c

[SYCL] fix no file in win rel (#6314)

NeoZhangJianyu committed 2 years ago

Verified a4f569e8

wpm : portable unicode tolower (#6305)

cebtenzzre committed 2 years ago

Verified 32c8486e

llama : greatly reduce output buffer memory usage (#6122)

compilade committed 2 years ago

Verified 557410b8

IQ1_M: 1.75 bpw quantization (#6302)

ikawrakow committed 2 years ago

Verified 55c1b2a3

convert-hf : fix exception in sentencepiece with added tokens (#6320)

pcuenca committed 2 years ago

Verified e097633f

quantize : be able to override metadata by key (#6321)

ikawrakow committed 2 years ago

Verified d25b1c31

embedding : adjust `n_ubatch` value (#6296)

mscheong01 committed 2 years ago

Verified deb72401

server : add `n_discard` parameter (#6300)

kaetemi committed 2 years ago

Verified 3d032ece

nix: make `xcrun` visible in Nix sandbox for precompiling Metal shaders (#6118)

josephst committed 2 years ago

Verified e190f1fc

cuda : rename build flag to LLAMA_CUDA (#6299)

slaren committed 2 years ago

Verified 28034596

nix: fix blas support (#6281)

ck3d committed 2 years ago

Verified b06c16ef

tests : include IQ2_XXS and IQ2_XS in test-quantize-fns (#6303)

ikawrakow committed 2 years ago

Verified 1f2fd4e7

flake.lock: Update (#6266)

ggerganov committed 2 years ago

Verified 43139cc5

cuda : fix LLAMA_CUDA_F16 build (#6298)

slaren committed 2 years ago

Verified 2f34b865

cuda : refactor into multiple files (#6269)

slaren committed 2 years ago

Verified ae1f211c

Server: clean up OAI params parsing function (#6284)

ngxson committed 2 years ago

Verified ad3a0505

[SYCL] fix SYCL backend build on windows is break by LOG() error (#6290)

NeoZhangJianyu committed 2 years ago

Verified 95ad616c

Older