OrpoLlama-3-8B / README.md
mayacinka's picture
Update README.md
c4d4455 verified
metadata
library_name: transformers
tags:
  - orpo
  - llama
  - llama 3
  - fine tune
license: apache-2.0
datasets:
  - mlabonne/orpo-dpo-mix-40k

Model Card for Model ID

Quick Llama 3 8B finetune with ORPO. Demontration that it can be fine tune in 2 hours only. Thanks to Maxime Labonne's notebook:

https://colab.research.google.com/drive/1eHNWg9gnaXErdAa8_mcvjMupbSS6rDvi?usp=sharing

  • Number of training samples from the dataset: 1500 out of 40K
  • Hardware Type: L4
  • Hours of training: 2
  • Cloud Provider: google colab