ngxson/llama.cpp

Pull Requests Commits
pass cpu-architecture arguments only to host code (C;C++) (#4943)
ngc92 committed 2 years ago
Verified 3e5ca793
llama : apply classifier-free guidance to logits directly (#4951)
dfriehs committed 2 years ago
Verified 44833967
awq-py : fix typo in awq-py/README.md (#4947)
ziliangpeng committed 2 years ago
Verified d9aa4ffa
cuda : fix dequantize kernel names (#4938)
ggerganov committed 2 years ago
Verified ddb008d8
llama : check for 256 divisibility for IQ2_XS, IQ2_XXS (#4950)
ikawrakow committed 2 years ago
Verified 2faaef39
CUDA: faster dequantize kernels for Q4_0 and Q4_1 (#4938)
ikawrakow committed 2 years ago
Verified 4a3156de
llama : fix missing quotes (#4937)
David Pflug committed 2 years ago
Verified a836c8f5
Add ability to use importance matrix for all k-quants (#4930)
ikawrakow committed 2 years ago
Verified 467a882f
llama : check LLAMA_TRACE env for extra logging (#4929)
ggerganov committed 2 years ago
Verified bb0c1392
scripts : sync-ggml-am.sh option to skip commits
ggerganov committed 2 years ago
Verified 9408cfda
llama : use LLAMA_LOG_ macros for logging
ggerganov committed 2 years ago
Verified 03c52674
Fix ffn_down quantization mix for MoE models (#4927)
ikawrakow committed 2 years ago
Verified a128c38d
metal : correctly set SIMD support flags on iOS (#4923)
azarovalex committed 2 years ago
Verified 5f5fe1bd
llama : support WinXP build with MinGW 8.1.0 (#3419)
guilt committed 2 years ago
Verified ac32902a
2-bit quantizations (#4897)
ikawrakow committed 2 years ago
Verified 147b17ac
Make Q3_K_S be the same as olf Q3_K_L for Mixtral-8x7B (#4906)
ikawrakow committed 2 years ago
Verified 807179ec
sync : ggml
ggerganov committed 2 years ago
Verified 76484fbf
ggml: cache sin/cos for RoPE (#4908)
JohannesGaessler committed 2 years ago
Verified c71d608c
metal : remove old API (#4919)
ggerganov committed 2 years ago
Verified 4be5ef55
server : fix prompt caching with system prompt (#4914)
ggerganov committed 2 years ago
Verified 0ea069b8
llama : fix detokenization of non-special added-tokens (#4916)
ggerganov committed 2 years ago
Verified f172de03
metal : disable log for loaded kernels (#4794)
ggerganov committed 2 years ago
Verified 2d57de52
llama : minimize size used for state save/load (#4820)
dfriehs committed 2 years ago
Verified df845cc9
workflows: unbreak nix-build-aarch64, and split it out (#4915)
SomeoneSerge committed 2 years ago
Verified 6b48ed08
main : add parameter --no-display-prompt (#4541)
YannFollet committed 2 years ago
Verified 722d33f3
gguf : fix potential infinite for-loop (#4600)
texmex76 committed 2 years ago
Verified c30b1ef3
metal : refactor kernel loading code (#4794)
ggerganov committed 2 years ago
Verified b38b5e93
compare-llama-bench: tweak output format (#4910)
JohannesGaessler committed 2 years ago
Verified 7dc78764
server : fix deadlock that occurs in multi-prompt scenarios (#4905)
ziedbha committed 2 years ago
Verified 356327fe
server : fix crash with multimodal models without BOS token (#4904)
makomk committed 2 years ago
Verified ee8243ad
Terms Of Use Privacy Policy Refund Policy Impressum