Add AG2 multi-agent system recipe#349
Conversation
New notebook demonstrating AG2 (formerly AutoGen) multi-agent framework with HF Inference API: basic two-agent chat, decorator-based tool use, and GroupChat orchestration with Researcher/Writer/Critic agents.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
Hi @merveenoyan @stevhliu ! Would love your review on this new AG2 multi-agent recipe when you get a chance. Happy to address any feedback! |
|
hi, thanks for your interest in contributing a recipe! currently, it feels a bit thin and branded and we would love to see if you could make it a bit more applied for specific practical use-cases |
Addresses reviewer feedback — more applied, less branded: - Focus on practical use case: automated model comparison pipeline - Scout/Analyst/Advisor agents with real HF Hub tools - Three tools: search_models, get_model_details, compare_model_stats - Interactive "Try It Yourself" section - Removed AG2 stats/branding, introduced naturally
faridun-ag2
force-pushed
the
add-ag2-multiagent-recipe
branch
from
7f532a9 to
7ddbd38
Compare
|
Hey @stevhliu, thanks for the feedback! Totally fair points. I've reworked the notebook — instead of a generic API walkthrough, it now builds a model comparison pipeline that actually solves a real problem: "which model should I pick from the Hub?" Three agents (Scout, Analyst, Advisor) use huggingface_hub tools to search models, read model cards, and compare stats before giving a recommendation. Also added a "Try It Yourself" section so readers can plug in their own use case. Stripped out the AG2 branding/stats — it's just mentioned naturally as the tool being used. Let me know what you think! |
2 participants
New Recipe: Building Multi-Agent Systems with AG2 and Hugging Face
Adds a notebook demonstrating AG2 (formerly AutoGen), an open-source multi-agent framework with 500K+ monthly PyPI downloads.
Content
The notebook covers three progressive examples:
Why this recipe
The cookbook has multi-agent examples with smolagents but none with AG2, one of the most popular multi-agent frameworks. AG2 takes a different approach — declarative agent definitions with automatic orchestration via GroupChat — which complements existing recipes.
Integration
AG2 connects to Hugging Face models via the OpenAI-compatible Inference API endpoint (
router.huggingface.co/v1), requiring only a HF token.Checklist
_toctree.ymlandindex.mdupdated