ggerganov/llama.cpp

Pull Requests Commits

sched : support async weight copy

slaren committed 1 year ago

5de9b743

readme : remove stray double quote (#7310)

danbev committed 1 year ago

Verified 8f7080bf

ggml : use dynamic thread scheduling for matrix multiplication (#6915)

kunnis committed 1 year ago

Verified e1b40ac3

Avoid unnecessarily disabling CUDA graphs (#7302)

agray3 committed 1 year ago

Verified dc020985

ggml : tag ggml_tensor::backend as deprecated (#7290)

slaren committed 1 year ago

Verified 344f9126

Add missing " (#7303)

AidanBeltonS committed 1 year ago

Verified 9a17ab91

embedding : free the batch after execution (#7297)

dm4 committed 1 year ago

Verified ea3b0590

ggerganov committed 1 year ago

Verified 29499bb5

ggml : add `ggml_upscale_ext` (ggml/814)

balisujohn committed 1 year ago

Verified 48aa8fd1

server bench: fix bench not waiting for model load (#7284)

JohannesGaessler committed 1 year ago

Verified 583fd6b0

script : sync ggml-rpc

ggerganov committed 1 year ago

Verified 9f773486

metal : support FA without mask + add asserts (#7278)

ggerganov committed 1 year ago

Verified e8a7fd4f

ggerganov committed 1 year ago

a5e3fde8

metal : tune soft_max number of threads (whisper/0)

ggerganov committed 1 year ago

f308ea70

ggml : try fix ppc64 (whisper/0)

ggerganov committed 1 year ago

c3c88f29

ggml : expose SSE3 and SSSE3 for MSVC when AVX is available (whisper/2128)

przemoc committed 1 year ago

182adefc

ggml : optimize for ppc64le using VSX intrinsics (ggml/784)

penghongbo committed 1 year ago

0d26d8cc

server: free sampling contexts on exit (#7264)

stevegrubb committed 1 year ago

Verified 4f026363

Revert "move ndk code to a new library (#6951)" (#7282)

mofosyne committed 1 year ago

Verified 1265c670

ggml : add RPC backend (#6829)

rgerganov committed 1 year ago

Verified 5e31828d

llama : disable pipeline parallelism with nkvo (#7265)

slaren committed 1 year ago

Verified 54160020

move ndk code to a new library (#6951)

eltonkola committed 1 year ago

Verified efc8f767

Add left recursion check: quit early instead of going into an infinite loop (#7083)

nuchi committed 1 year ago

Verified e0f55618

docs: Fix typo and update description for --embeddings flag (#7026)

louixs committed 1 year ago

Verified 27f65d62

convert-hf : support direct Q8_0 conversion (#7234)

compilade committed 1 year ago

Verified ee522250

llama : less KV padding when FA is off (#7257)

ggerganov committed 1 year ago

Verified 614d3b91

llava-cli: fix base64 prompt (#7248)

Adriankhl committed 1 year ago

Verified 30e70334

perplexity: add BF16 vs. FP16 results (#7150)

JohannesGaessler committed 1 year ago

Verified 1c570d8b

[SYCL] rm wait() (#7233)

arthw committed 1 year ago

Verified 948f4ec7

llama : rename jina tokenizers to v2 (#7249)

JoanFM committed 1 year ago

Verified 9aa67249

Older