Gary3410

Datasets

4

In the paper, you select a portion of the DOTA dataset for training and validation, can you share it?

Issues with fine-tuning on the ScienceQA dataset.

4

I have some problems fine-tuning llama_adapter on the ScienceQA dataset. I'm not quite sure how to write a prompt on the ScienceQA dataset according to the current template. Is it...

Alpaca finetuning issues

3

Thanks for your wonderful work！ I had a problem when fine-tuning the model. https://github.com/ZrrSkywalker/LLaMA-Adapter/blob/5f1b37e0e2f3ab2e423ea71234c89829fa271ad7/alpaca_finetuning_v1/llama/model.py#L80-L85 https://github.com/ZrrSkywalker/LLaMA-Adapter/blob/5f1b37e0e2f3ab2e423ea71234c89829fa271ad7/llama/model.py#L78-L83 The `self.n_local_heads` in training and inference are not the same, will this affect the deployment...

How to train with multiple GPUs？

1

Hi! Thank you for this great job in open source! I'm currently reproducing single-task training on an RTX3090, but the training time costs are high. Can the training script now...