Augmenting pre trained models with Bumblebee

Elixir Framework Forums Nx Forum
1

This blogpost says:

Next, we plan to focus on training and transfer learning of Neural Networks in Elixir, allowing developers to augment and specialize pre-trained models according to the needs of their businesses and applications.

I have been trying bumblebee out and I am really impressed. I am since looking forward to the feature of augmenting pre trained models with own data. Has there been any news on this ?

2

There are some guides and I know some folks on the #machine-learning channel of the Erlang Ecosystem Foundation Slack have played with this but this is not as automated as we want it to be.

3

Thank you for the response. I am a complete beginner regarding AI and I am approaching the topic from the Elixir side. I didn’t even know of hugging face before your video about Bumblebee and GPT2 :smile:. I will have a look at the Erlang Slack.

Does it mean the plans of the Bumblebee team changed and that we should not expect a solution for augmenting pre trained models ?

4

We will get back to it eventually. It is just that we are a small team. :slight_smile:

3 Likes
5

Great to hear ! Your productivity and quality of work makes it seem like you have a big team : ). I always think, when dealing with other ecosystems, if only José Valim would be the lead here, they would be miles ahead and I would not be fighting all these bugs and issues.

6

@josevalim are there any new updates / hints on this ?

7

Jumped the gun too early, this might be a starting point. Fine-tuning — Bumblebee v0.6.0 for anyone who gets here :slight_smile: