3 ms·
I would like to support dreambooth, if there is a way to store the fine-tuned models more efficiently. The challenge is that each trained model is quite large a
by tehsauce 4y ago
I would like to support dreambooth, if there is a way to store the fine-tuned models more efficiently. The challenge is that each trained model is quite large and a bunch of models can't fit into one gpus memory at once.
- conidig 4y agotextual inversion maybe? lightweight embeddings and I haven't seen any API offering it at the moment.
- spyder 4y agoTry the Lora fine-tuning method with few MB results: https://github.com/cloneofsimo/lora https://github.com/cloneofsimo/lora
- tehsauce 4y agoThis looks great! Thanks for sharing.