ggerganov/llama.cpp

Pull Requests Commits

imatrix: guard Q4_0/Q5_0 against ffn_down craziness

Iwan Kawrakow committed 1 year ago

bb9abb5c

imatrix: adding support for legacy quants

Iwan Kawrakow committed 1 year ago

6f9ec42a

pass cpu-architecture arguments only to host code (C;C++) (#4943)

ngc92 committed 1 year ago

Verified 3e5ca793

llama : apply classifier-free guidance to logits directly (#4951)

dfriehs committed 1 year ago

Verified 44833967

awq-py : fix typo in awq-py/README.md (#4947)

ziliangpeng committed 1 year ago

Verified d9aa4ffa

cuda : fix dequantize kernel names (#4938)

ggerganov committed 1 year ago

Verified ddb008d8

llama : check for 256 divisibility for IQ2_XS, IQ2_XXS (#4950)

ikawrakow committed 1 year ago

Verified 2faaef39

CUDA: faster dequantize kernels for Q4_0 and Q4_1 (#4938)

ikawrakow committed 1 year ago

Verified 4a3156de

llama : fix missing quotes (#4937)

David Pflug committed 1 year ago

Verified a836c8f5

Add ability to use importance matrix for all k-quants (#4930)

ikawrakow committed 1 year ago

Verified 467a882f

llama : check LLAMA_TRACE env for extra logging (#4929)

ggerganov committed 1 year ago

Verified bb0c1392

scripts : sync-ggml-am.sh option to skip commits

ggerganov committed 1 year ago

Verified 9408cfda

llama : use LLAMA_LOG_ macros for logging

ggerganov committed 1 year ago

Verified 03c52674

Fix ffn_down quantization mix for MoE models (#4927)

ikawrakow committed 1 year ago

Verified a128c38d

metal : correctly set SIMD support flags on iOS (#4923)

azarovalex committed 1 year ago

Verified 5f5fe1bd

llama : support WinXP build with MinGW 8.1.0 (#3419)

guilt committed 1 year ago

Verified ac32902a

2-bit quantizations (#4897)

ikawrakow committed 1 year ago

Verified 147b17ac

Make Q3_K_S be the same as olf Q3_K_L for Mixtral-8x7B (#4906)

ikawrakow committed 1 year ago

Verified 807179ec

ggerganov committed 1 year ago

Verified 76484fbf

ggml: cache sin/cos for RoPE (#4908)

JohannesGaessler committed 1 year ago

Verified c71d608c

metal : remove old API (#4919)

ggerganov committed 1 year ago

Verified 4be5ef55

server : fix prompt caching with system prompt (#4914)

ggerganov committed 1 year ago

Verified 0ea069b8

llama : fix detokenization of non-special added-tokens (#4916)

ggerganov committed 1 year ago

Verified f172de03

metal : disable log for loaded kernels (#4794)

ggerganov committed 1 year ago

Verified 2d57de52

llama : minimize size used for state save/load (#4820)

dfriehs committed 1 year ago

Verified df845cc9

workflows: unbreak nix-build-aarch64, and split it out (#4915)

SomeoneSerge committed 1 year ago

Verified 6b48ed08

main : add parameter --no-display-prompt (#4541)

YannFollet committed 1 year ago

Verified 722d33f3

gguf : fix potential infinite for-loop (#4600)

texmex76 committed 1 year ago

Verified c30b1ef3

metal : refactor kernel loading code (#4794)

ggerganov committed 1 year ago

Verified b38b5e93

compare-llama-bench: tweak output format (#4910)

JohannesGaessler committed 1 year ago

Verified 7dc78764

Older