This repository contains minimal code to run models from Nyonic Model Factory which produces a set of large language models with a stable and fast data processing and training pipeline.
Latest
04.25.2024 - nyonic releases wonton-7B (base) model for research and use, trained by nyonic’s model training infrastructure. Request access to wonton-7B.
03.05.2024 – nyonic releases wonton-3B (base) model for research and use, trained by nyonic’s model training infrastructure
02.01.2024 – nyonic releases wonton-1.5B (base) model for research and use, trained by nyonic’s model training infrastructure
git clone https://github.com/nyonicai/nyonic-public.git
cd nyonic-public
pip install -r requirements.txt
Please fill in this form to request access to our model(s). The model file shoud be placed under the models folder.
For a trivial test, simply run
python -m main
which use default inference settings from wonton-6.7B, you can specify a model you want use with --model_conf argument, or change any supported sampling parameters. A complete example:
python -m main --model_conf /path/to/model_conf.yaml --max_tokens 200 --strategy top_p --top_p 0.6 --temperature 0.8 --device cuda:0