Anyone have a preferred library/technique for training draft models? I've found the FastDraft paper, and NVIDIA supports training EAGLE-3 models with their optimizer library.

I want to accelerate Mistral 3 Small (make a 2-3B with the same vocab). There doesn't seem to be a good existing model.

This is a neat new variant on RAG - no vectors, not even full-text search, instead showing the model a header hierarchy and giving it a tool to read the relevant sections

My notes here: simonwillison.net/2024/Dec/6/r...

The default "Discover" feed is pretty bad, and the "What's Hot (Classic)" is worse- just a lot of low effort content, click bait, and lowest common denominator posts. Are there guides on how to improve this? Will the recommendations improve slowly with increased interaction/relevance feedback?

I think more people should evangelize the old spirit of the Internet - open source, open platforms, open data. If email didn't already exist and it was created today, it'd be locked in proprietary AmaGooMetaSoft gardens.

Looking for ML/NLP researchers to follow, busy going through my following list from elsewhere. Hello world! 👋