MAGAer13

We would not update the paper but we will include the specification of the model's design in the video branch. The code and weight's will be released

> > > really need it ! > > > > > > We will release the video version in this week! > > Hi, did you release the video...

See #101. Also we have update the checkpoint in HF.

Nice suggestion, you only need to have a GPU with **24GB memory (i.e. RTX 3090) or EVEN 16GB memory (i.e., V100)** under **fp16 or bf16 precision**. We update in the...

You can just add your options into prompt, and use as an open-generation style. We will release mPLUG-Owl-2 recently, which is a better foundation model, and it can better support...

The xxx_sft is just the indicator of dataset name and task type. You just need to specify xxx as your own dataset name

We random mix the text data and multi-modal data. For each batch, we do not control the ratio, it just random sampled, and the ratio within a batch would similar...

I think you miss the token as the placeholder for the image inputs? You may try this: ``` {"image": ["image1.jpg","image2.jpg"], "text": "The following is a conversation between a curious human...

I have met the similar case. I think there are some overflow during the training. I recommend you to have a look on the validation which is more reliable.

We have not tested 4bit.