ggerganov/llama.cpp

Pull Requests Commits

Merge and update

SlyEcho committed 2 years ago

Verified 28046d1e

CUDA: faster non k-quant mul_mat_q kernels (#2483)

JohannesGaessler committed 2 years ago

Verified 468ea24f

CUDA: Fix models with output size != 32000 (#2480)

JohannesGaessler committed 2 years ago

Verified 4f6b60c7

readme : add Aquila-7B model series to supported models (#2487)

ftgreat committed 2 years ago

Verified 220d9318

tests : Fix compilation warnings (Linux/GCC) (#2451)

netrunnereve committed 2 years ago

Verified 81844fbc

readme : Add Chinese LLaMA-2 / Alpaca-2 to supported models (#2475)

ymcui committed 2 years ago

Verified a312193e

fix a typo in examples/server/README.md (#2478)

lvscar committed 2 years ago

Verified c574bddb

server : Support dark mode (#2414)

ebraminio committed 2 years ago

Verified 86aeb277

metal : add gqa8 kernel to allow llama-2-70B on metal (#2459)

mbosc committed 2 years ago

Verified 1873ff58

CUDA: fixed LLAMA_FAST compilation option (#2473)

JohannesGaessler committed 2 years ago

Verified 49e7cb5b

CUDA: fixed cmake F16 option (#2471)

JohannesGaessler committed 2 years ago

Verified b772bba4

CUDA: mmq CLI option, fixed mmq build issues (#2453)

JohannesGaessler committed 2 years ago

Verified 0728c5a8

CUDA: Implemented row flattening for non-glm RoPE (#2468)

JohannesGaessler committed 2 years ago

Verified 1215ed7d

CUDA: fewer memory bank conflicts for mul_mat_q (#2458)

JohannesGaessler committed 2 years ago

Verified 2dbf5189

Fix Metal backend broken from the allocator changes (#2455)

slaren committed 2 years ago

Verified 9d2382b3

ggml : add graph tensor allocator (#2411)

slaren committed 2 years ago

Verified a1136895

CUDA: Quantized matrix matrix multiplication (#2160)

JohannesGaessler committed 2 years ago

Verified 11f3ca06

CUDA: faster multi GPU synchronization (#2448)

JohannesGaessler committed 2 years ago

Verified 9baf9ef3

perplexity : add Hellaswag calculation (#2389)

klosax committed 2 years ago

Verified 8a88e585

ggml : workaround for missing _mm256_setr_m128i in GCC < 8 in k_quants.c (#2405)

lx200916 committed 2 years ago

Verified a9559bf7

llama : support more diverse tokenizers? (#2420)

eric8607242 committed 2 years ago

Verified ee1b497c

examples : fix whitespace

ggerganov committed 2 years ago

Verified d73b8d48

examples : server chat mode with llama2 (#2400)

nhamanasu committed 2 years ago

Verified 34ae1caf

readme : fix the description of the Tail free sampling (TFS) method (#2431)

WeirdConstructor committed 2 years ago

Verified d91f3f0c

llama : use n_embd_gqa instead of n_embd to handle llama-2 70B (#2433)

randxie committed 2 years ago

Verified 65cdf34b

Obtaining LLaMA 2 instructions (#2308)

niansa committed 2 years ago

Verified edcc7ae7

convert.py : Update to support 70B HF format model files (#2427)

mj-shifu committed 2 years ago

Verified 7c529ced

metal : disable graph concurrency optimization due to bug (#2413)

ggerganov committed 2 years ago

Verified 1a941869

ggml : fix assert in ggml_set_unary_op (#2410)

slaren committed 2 years ago

Verified b5472ea0

make : build with -Wmissing-prototypes (#2394)

cebtenzzre committed 2 years ago

Verified 6df1f594

Older