For months, we've worked on building @huggingface's new moonshot: agentic systems. So today we're very proud to announce the release of 𝚜𝚖𝚘𝚕𝚊𝚐𝚎𝚗𝚝𝚜! It's the simplest library we could make to let people build powerful agents. 💥 The main logic for agents fits in ~1000 lines of code. So it's really dead simple. 🧑‍💻 The main agent class is CodeAgent, and agent that writes its actions in code. That means, contrary to the standard set by OpenAI of writing tool calls as JSON blobs, this agent writes code snippets. It's much more natural for LLMs to write actions this way, and as a result performance is vastly improved. 🌍 It supports any LLM through @LiteLLM integration. 🛡️ We enabled secure code execution via @e2b_dev sandboxes.

@paulg Study here:

Here's the blog post:

@ADarmouni Thank you @ADarmouni ! Here's the blog post:

RT @mervenoyann: free the agents @AymericRoucher @avillanovamoral @huggingface 🕵🏻🤗

@huggingface @OpenAI The demo is flooded under requests at the moment. To skip the queue, you can duplicate the space under your own Hub account; but it will still require providing your own API keys as space Secret variables!

Please duplicate the Space, at the moment it's completely underwater with all the requests!

@NaanLeCun @huggingface @OpenAI That would be a good plan! I should have asked Santa to gift me 24 more hours in a day 😅

@peker_eth @huggingface @OpenAI It's really fast, but performs less well than o1 and gpt-4o. I guess models too small don't cut it yet for hard tasks.

@KimNoel399 @huggingface @OpenAI My previous submission, months ago with transformers.agents on GPT-4o scored a bit over 40% ont the same benchmark. We've been a long way since then with smolagents!

@bevenky @huggingface @OpenAI Yes but didn't work as well as o1. It was not a dumbness issue as with many LLMs, more a lack of adaptation to the the framework's guidelines. So we're contemplating fine-tuning to solve this !

Pick the best global source of truth for humanity: - Wikipedia, transparently crowd-sourced from the knowledge of thousands of contributors - ChatGPT, has an opaque set of values ~chosen by @sama under incentives from the richest/more powerful influence group I would fear a world where one person can change global truth by turning a knob during post-training.

@sama With an agentic setup, quite obvious that scores would increase! The comparison against other agents on the GAIA benchmark is more interesting.

@russeurope @PascalJouary @russeurope à quelles élections faites vous référence pour le Royaume Uni? Je vois que les élections générales ont été repoussées en 45:

@pablosanzo @mervenoyann We did a blog post demoing an example of it, but it's the first release that makes it part of the core code (contributed by @merve)

@9Knowled9e @victormustar pip install smolagents And add [dev] to get the whole stuff!