Rupesh-rkgit / FineTuning-and-Inference-Llama2

Finetuning and Inference of Llama2 7b model on colab
14 stars 0 forks source link

Fine Tuning and inference of LLama V2 7b on colab

To get access to the models, first fill up the form here with the same mail id used to create huggingface account. 

Link : https://ai.meta.com/resources/models-and-libraries/llama-downloads/

Huggingface Model card : https://huggingface.co/meta-llama

Dataset : https://huggingface.co/datasets/timdettmers/openassistant-guanaco

Llama Model size

Llama Model size