Difference between show-1-base-0.0 show-1-base?

[zhizhou57]

When I download show-1 model from huggingface, I found there are two repository: showlab/show-1-base-0.0 and showlab/show-1-base. It seems that their text_encoder have different torch_dtype. For show-1-base, torch_dtype is float32, and for show-1-base-0.0, torch_dtype is float16. Is there other differences between these two repository? And what are these model used for respectively？
Btw I encounter the type error when encode the prompt, expected scalar type Float but found Half

[junhaozhang98]

(1)unet of base-0.0 is trained on webvid-10M with watermark

(2)base is trained on webvid-10M and finetuned on intervideod dataset without watermark.

Since webvid has a better caption quality, base-0.0 may have better text-video alignment.