w4nderlust

5 ms·

1.
by w4nderlust 2y ago
We got very similar findings: we published a paper that show that smaller LLMs (3-7b) when finetuned with LoRA can match or outperform GPT-4 on a variety of tasks (29 out of 31) including classification, summarization, info extraction, &quo
2.
by w4nderlust 2y ago
Predibase ( http://predibase.com ), also referred in the article, is a platform specifically designed for exactly that. It also has "repos" for finetuning multiple models and comapre their performance and keeping thing
3.
by w4nderlust 2y ago
Here is an example of the Predibase platform, referred in the article for the Solar model, but that can train also Llama-3, Phi-3 and Mistral. https://www.youtube.com/watch?v=R2JQhzfaOFw&themeRefresh=1 I think you can a
4.
Predibase: the enterprise declarative machine learning platform(medium.com)
10 points by w4nderlust 4y ago|0 comments
5.
Ludwig v0.3 Released(eng.uber.com)
1 points by w4nderlust 6y ago|0 comments
6.
by w4nderlust 8y ago
Speech is coming to Ludwig soon, we are currently working on that.
7.
by w4nderlust 8y ago
You may be right about that, but also it depends on the requirements that you have. Ludwig gives you a lot of options for those tricks, like for instance gradient clipping or regularizers, or learning rate and batch size scheduling, but tho
8.
by w4nderlust 8y ago
Text is just one of the possible types of features supported in Ludwig. For those features, no, you can train models on any language, with a couple little caveats: you can train both character based models and word level ones. For character
9.
by w4nderlust 8y ago
That experience you talked about, being able to really quickly come up with solutions starting from an already established and working configuration, was one of the main motivating factors behind the making of Ludwig too! Basically I had a
10.
by w4nderlust 8y ago
Your configuration file looks really good! Why did you stop pursuing it? To your points: * That may be true, but I think it also depends on the level of abstraction. Fon instance in Ludwig the encoder, although configurable, are pretty mono
11.
by w4nderlust 8y ago
You have a good point about debugging. So far my approach has been to write quite long and explanatory error messages (like this https://github.com/uber/ludwig/blob/9c9e5de56dcad89461c5d2ec... for instance) a
12.
by w4nderlust 8y ago
We are considering a bunch of options re feature store. I personally like EuclidDB, but I'm not sure it fits the notion of feature store that we need (while it's great for ANN inference). Anyway we are looking into it. Regarding t
13.
Introducing Ludwig, a Code-Free Deep Learning Toolbox(ubere.ng)
1 points by w4nderlust 8y ago|0 comments
14.
by w4nderlust 9y ago
I had the chance to read a preprint, it's a great survey, it will be published soon.
15.
by w4nderlust 9y ago
I edited the page. Now there's a download button and a preview button. If you click the preview button, the javascript document viewer embedding the pdf will be loaded, otherwise it won't be loaded.