Hugging Face's logo

NumbersStation
/
nsql-350M

Text Generation
Transformers
PyTorch
codegen
Model card Files
xet
 Community
8

Few questions from the blog

#3
by snehilsanyal - opened
Discussion
snehilsanyal

Hey NumbersStation team, congratulations for the work, finally an "open-source" FM specialized for text to sql conversion is available for the community due to your efforts. I had a few questions from the blog,

  1. Any plans for a research paper?
  2. Will the fine-tuning script be made available? From the blog, companies can use their own company data to enhance the NSQL models, so an example would be better.
  3. Any suggestions for the deployment and serving of these FMs? Especially the 6B model?
senwu
 NumbersStation org

Hi @snehilsanyal ,

Thanks for your interest in our work! Please find the response to your questions below:

  1. Any plans for a research paper?

    We have a plan for that and we will share more insights of our text-to-SQL work.

  2. Will the fine-tuning script be made available? From the blog, companies can use their own company data to enhance the NSQL models, so an example would be better.

    We don't have a plan to release our fine-tuning script right now but we have released the text-to-SQL data we collected here along with NSQL models.

Any suggestions for the deployment and serving of these FMs? Especially the 6B model?

Deploying the 6B model requires around 16G GPU memory, so a 24G GPU should be efficient to deploy and serve the model.
snehilsanyal

Thank you @senwu for your reply.
The blog also mentioned that a future direction for this would be from "Textbooks are all you need" paper, which is a really interesting direction for the work. Excited to see how it goes :D
Many congratulations on the work

Sign up or log in to comment