Welcome to TexTeller Discussions!

[OleehyO]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[YogeLiu]

Hello, this work is very greatful. I want to know if you opento the dataset? Or if can provide the detail about the dataset,for example the class(math: matrix, simple formula, complex formula, linear algebra, etc.)

[OleehyO]

Hello, the dataset will be open-sourced after the paper is published, which we expect to be after August. The data has been crawled from arXiv, so it includes a variety of formula types.

[YogeLiu]

Thanks for your reply.
I get. Will you continue train the model in the future? I found this version model has issue for some complex formula, for example, mutli-line matrix.

[OleehyO]

Yes, we will consider further increasing the amount of data and extending the context window length in the future. However, you can first try fine-tuning using images of pure formulas to see if better results can be achieved.

[YogeLiu]

ok, I see. Can you share the training detail, how long it takes and the configure of GPU? If the paper is publish, can share the doi or url?

[OleehyO]

The link will be posted on the project homepage after the paper is published.

Regarding the training configuration, TexTeller 3.0 was trained on an A800 for 2,688 hours.