A side-by-side comparison of release history, capabilities, licensing and model sizes.
| Attribute |
 FLAN-T5
|
 Vicuna
|
|---|---|---|
| Developer | UC Berkeley, CMU, Stanford, MBZUAI, UCSD | |
| Current generation | – | – |
| Initial release | 2022-12-06 | 2023-03-30 |
| Instruct tuned | Yes | Yes |
| Coding | No | No |
| Finetuning | Yes | Yes |
| Open source | Yes | Yes |
| License | Apache 2.0 | Noncommercial |
| Model sizes | 3B, 11B | 13B |
| Links |
FLAN-T5 is a finetuned version of Google's popular T5 model with instruct-finetuning. As stated in the model repository's introduction, compared to T5, FLAN-T5 is "just better at everything." With its permissive license, FLAN-T5 has become a popular option for a starting instruct model.
Released alongside Koala, Vicuna is one of many descendants of the Meta LLaMA model trained on dialogue data collected from the ShareGPT website. According to the authors, Vicuna achieves more than 90% of ChatGPT's quality in user preference tests, while vastly outperforming Alpaca. As of May 2023, Vicuna seems to be the heir apparent of the instruct-finetuned LLaMA model family, though it is also restricted from commercial use.
