You can just ask things 🗣️

"show me messages in the coding category that are in the top 10% of reward model scores"

Download really high quality instructions from the Argilla Llama3.1 405B synthetic dataset 🔥

It doesn't get easier than this. Why are you writing SQL by yourself when it's almost 2025

The amazing, new Qwen2.5-Coder 32B model can now write SQL for any @hf.co dataset ✨

This is insane! Structured generation in the browser with the new @hf.co SmolLM2-1.7B model

• Tiny 1.7B LLM running at 88 tokens / second ⚡
• Powered by MLC/WebLLM on WebGPU 🔥
• JSON Structured Generation entirely in the browser 🤏

You can literally do the histogram in one line in less than 10 seconds 💨

> from histogram(train, "Average ⬆️")

Here's what the model licenses look like:

Lots of great open licenses in there too! 💪

The OpenLLM Leaderboard just passed 2k evals 🥳

Here's a look at the distribution of average scores for all those models!

Great work by the @huggingface.bsky.social team to do these evals!

** log and get out of the way **

Automatically tracking all Ollama requests to a dataset with the new observers python library!

With just a few lines of code all your requests can be sent to @huggingface.bsky.social datasets for annotating, analysis and observability 🔭

The main three stores are:
• DuckDB (local, SQL over traces)
• Hugging Face Datasets (dataset viewer, sql console)
• Argilla - annotation and filtering UI

observers 🔭 - automatically log all OpenAI compatible requests to a dataset 💽

• supports any OpenAI compatible endpoint 💪
• supports @duckdb.org, @huggingface.bsky.social datasets and Argilla as stores

> pip install observers

Now they do!
t.co/T1WhhBIAqS

quick to implement it too