ggerganov/llama.cpp

Pull Requests Commits

gguf : start write tensor info

monatis committed 1 year ago

af1c9966

refactor: reduce code duplication and better API

monatis committed 1 year ago

8332d261

gguf : add comments

ggerganov committed 1 year ago

Verified d8491fc7

gguf : read / write sample models

ggerganov committed 1 year ago

Verified 5628ec71

gguf : gguf.c is now part of ggml.c

ggerganov committed 1 year ago

Verified e46870f5

gguf : simplify gguf_get_val

ggerganov committed 1 year ago

Verified d313c0fa

gguf : do not support passing existing ggml_context to gguf_init

ggerganov committed 1 year ago

Verified cb871fa0

gguf : add gguf_get_tensor_name()

ggerganov committed 1 year ago

Verified 860c9c63

gguf : initial model loading - not tested

ggerganov committed 1 year ago

Verified 78b226a9

gguf : read tensor info

ggerganov committed 1 year ago

Verified d91b985d

gguf : read header + meta data

ggerganov committed 1 year ago

Verified 8d6acfec

gguf : first API pass

ggerganov committed 1 year ago

Verified 68731487

ci : disable CI temporary to not waste energy

ggerganov committed 1 year ago

Verified 7e82d25f

wip : implement GGUF (#2397)

monatis committed 1 year ago

Verified bae6b125

ggerganov committed 1 year ago

Verified 4d698495

ggml : allocate graphs in a context (#2392)

slaren committed 1 year ago

Verified 5488fb78

Add LLAMA_DEFAULT_RMS_EPS so we can change the default (#2384)

ikawrakow committed 1 year ago

Verified eb542d39

ggml : fix ggml_flash_attn to use op_params (#2387)

slaren committed 1 year ago

Verified 07aaa0f6

convert.py : support bpe tokenizer (#2228)

ftgreat committed 1 year ago

Verified fce48caf

ggml : relax contiguous constraints in activation function (#2371)

li-plus committed 1 year ago

Verified 875086bd

ggml : improve graph build time via hash table lookup (#2329)

slaren committed 1 year ago

Verified da188983

build : fix line breaking error in build-info.sh (#2349)

hesenp committed 1 year ago

Verified 82552b7f

main : add `--in-prefix-bos` to prefix BOS to user inputs; keep EOS (#2304)

jxy committed 1 year ago

Verified 0c06204f

ci : add non-AVX scalar build/test (#2356)

netrunnereve committed 1 year ago

Verified 1fed755b

k_quants : add AVX support to dot functions with QK_K as 64 (#2339)

katsu560 committed 1 year ago

Verified be2301bc

metal : concurrently dispatch commands (#2358)

lshzh-ww committed 1 year ago

Verified 1aa18ef9

Another speed gain for Q4_0 and Q4_1 on Metal (#2375)

ikawrakow committed 1 year ago

Verified 9a08eaf3

Fix Q4_K and Q5_K for QK_K = 64 on CUDA (#2359)

ikawrakow committed 1 year ago

Verified 129d844c

server: add rms_norm_eps parameter (#2380)

slaren committed 1 year ago

Verified d5512b78

[Server] Escape HTML in webchat (#2368)

SlyEcho committed 1 year ago

Verified c798308e

Older