Closed mayhugotong closed 1 year ago
We did not use the publicly available Alpaca-LoRA model. If you are using it, please align the parameters related to LoRA in your script with the model you are using.
If you download the model from the huggingface, you could click the "Files and Versions" and choose the adapter_config.json, you will see the hyper-parameters for LoRA.
May I ask which Alpaca-LoRA weight you are using for your RT phase? Is it the Tleon one? There's model size mismatching error.