ggerganov/llama.cpp

Pull Requests Commits

cuda : reduce registers

ggerganov committed 1 year ago

Verified 4c190ba6

cuda : bump nwarps by 1

ggerganov committed 1 year ago

Verified 5dd355fe

cuda : adapt soft_max to F16 mask and pos

ggerganov committed 1 year ago

Verified 08e69c50

Merge branch 'master' into gg/flash-attn

ggerganov committed 1 year ago

Verified 3e318e76

metal : improve perf via smaller int registers

ggerganov committed 1 year ago

Verified 57c03b78

[SYCL] Revisited & updated SYCL build documentation (#6141)

OuadiElfarouki committed 1 year ago

Verified 5106ef48

convert : refactor vocab selection logic (#6355)

cebtenzzre committed 1 year ago

Verified be55134a

llava : fix MobileVLM (#6364)

ZiangWu-77 committed 1 year ago

Verified 66ba5602

llama : fix command-r inference when omitting outputs (#6367)

compilade committed 1 year ago

Verified 0308f5e3

ci: bench: fix master not schedule, fix commit status failed on external repo (#6365)

phymbert committed 1 year ago

Verified 28cb9a09

doc: fix outdated default value of batch size (#6336)

Sunt-ing committed 1 year ago

Verified cfc4d75d

server : stop gracefully on SIGTERM (#6348)

EZForever committed 1 year ago

Verified 6902cb7f

nix: removed unnessesary indentation

hutli committed 1 year ago

d2d8f389

nix: moved blas availability check to package inputs so it is still overridable

hutli committed 1 year ago

d39b308e

using blas.meta.available to check host platform

hutli committed 1 year ago

c8739766

only using explicit blas if hostPlatform is allowed

hutli committed 1 year ago

dbb03e2b

nix: .#windows: proper cross-compilation set-up

SomeoneSerge committed 1 year ago

e9f17dc3

nix: package: don't introduce the dependency on python

SomeoneSerge committed 1 year ago

22a462cc

nix: .#widnows: init

hutli committed 1 year ago

f6a0f5c6

doc: fix typo in MobileVLM-README.md (#6181)

ZiangWu-77 committed 1 year ago

Verified d0e2f641

[SYCL] fix set main gpu crash (#6339)

NeoZhangJianyu committed 1 year ago

Verified 25f4a613

server: continuous performance monitoring and PR comment (#6283)

phymbert committed 1 year ago

Verified a016026a

nix: ci: dont test cuda and rocm (for now)

SomeoneSerge committed 1 year ago

53c7ec53

ggml : fix bounds checking of zero size views (#6347)

slaren committed 1 year ago

Verified e5b89a44

make : whitespace

ggerganov committed 1 year ago

Verified 3a034597

embedding : show full embedding for single prompt (#6342)

howlger committed 1 year ago

Verified 1e13987f

cuda : fix build

ggerganov committed 1 year ago

Verified 6be02b59

Merge branch 'master' into gg/flash-attn

ggerganov committed 1 year ago

Verified 013721df

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

AidanBeltonS committed 1 year ago

Verified e82f9e2b

Make IQ1_M work for QK_K = 64 (#6327)

ikawrakow committed 1 year ago

Verified cbc83436

Older