Loading repository data…
tatwan / repository
This is a 6-module, hands-on mini-course for building AI agents using Hugging Face's smolagents library. It targets intermediate-to-advanced practitioners who are comfortable with Python and have basic familiarity with language models.
A transparent discovery signal based on current public GitHub metadata.
This score does not audit code, security, maintainers, documentation quality, or suitability. Verify the repository and its current documentation before adoption.
A practical, code-heavy mini-course for data engineers and ML practitioners.
This is a 6-module, hands-on mini-course for building AI agents using Hugging Face's smolagents library. It targets intermediate-to-advanced practitioners who are comfortable with Python and have basic familiarity with language models. Every module uses open-source tools and free-tier APIs — no paid subscriptions required to complete the core curriculum. The modules follow a linear progression, where each one builds directly on concepts and code introduced in the previous module.
| Requirement | Notes |
|---|---|
| Python 3.10+ | |
| uv (package manager) | Install: curl -LsSf https://astral.sh/uv/install.sh | sh |
| HuggingFace account | Free at huggingface.co — needed for HF_TOKEN |
| Basic Python proficiency | Comfortable with functions, classes, imports |
| Basic ML familiarity | Know what a language model is |
cd smolagentsuv sync — installs all dependenciescp .env.example .env — then edit .env and add your HF_TOKENuv run jupyter lab — opens Jupyter in your browser01_foundations/notebook.ipynbGet your HF_TOKEN at https://huggingface.co/settings/tokens — free account, read access is sufficient.
HF_TOKEN · Value: your HuggingFace token!uv pip install smolagents python-dotenv duckduckgo-search mlflow
## Setup cell and run it:
from google.colab import userdata
os.environ['HF_TOKEN'] = userdata.get('HF_TOKEN')
| Module | Topic | Key Tools | Open in Colab |
|---|---|---|---|
| 01 | Foundations: The Agent Loop | CodeAgent, InferenceClientModel |  |
| 02 | Tools & Custom Tools | @tool, Tool subclass | |
| 03 | CodeAgent vs ToolCallingAgent | Both agent types, LiteLLMModel | |
| 04 | Web Search & Browsing | DuckDuckGoSearchTool, VisitWebpageTool | |
| 05 | Multi-Agent Orchestration | managed_agents, manager pattern | |
| 06 | MLflow Observability | mlflow, step tracing, run comparison | |
Running in Colab? Each notebook has a commented-out cell at the top to install dependencies (
!uv pip install ...) and a section for loading yourHF_TOKENvia Colab Secrets. See the Quick Start for Colab section below.
Each module folder contains three files:
outline.md — what the module covers; intended as an instructor reference and planning documentnotebook.ipynb — the main learning artifact; this is what you run and work throughinstructions.md — module-specific prerequisites, step-by-step run instructions, and common errors with solutionsWork through the notebook.ipynb in each module in order. Refer to instructions.md if you hit a setup issue, and outline.md if you want to understand the full scope of a module before diving in.
| Tool | Purpose | Free? |
|---|---|---|
| smolagents | Agent framework | Yes (Apache 2.0) |
| HuggingFace Inference API | LLM backend | Yes (free tier) |
| DuckDuckGo Search | Web search for agents | Yes |
| MLflow | Experiment tracking & observability | Yes (Apache 2.0) |
| uv | Fast Python package manager | Yes |
| OpenAI API | Optional: Module 03 model-swap demo | Paid (optional) |
Q: Do I need a GPU?
No. All models run on the HuggingFace Inference API (free tier). No local GPU is required.
Q: Can I use a local model with Ollama instead of HuggingFace?
Yes. Module 03 shows how to swap model providers. Use:
LiteLLMModel(model_id="ollama_chat/llama3.2", api_base="http://localhost:11434")
Q: Can I use OpenAI or Anthropic?
Yes. Module 03 demonstrates the model swap. For example:
LiteLLMModel(model_id="gpt-4o-mini", api_key=os.environ["OPENAI_API_KEY"])
The rest of the agent code is unchanged — smolagents abstracts the model backend cleanly.
Q: The notebooks use Qwen/Qwen2.5-Coder-32B-Instruct — can I change the model?
Yes. Replace model_id in the setup cell of any notebook with any HuggingFace model that supports chat. Recommended alternatives:
meta-llama/Llama-3.3-70B-Instructmistralai/Mistral-7B-Instruct-v0.3Q: I'm getting rate limit errors from DuckDuckGo.
Add time.sleep(3) between agent runs, or reduce max_steps. DuckDuckGo's free API has a soft rate limit that can trigger under rapid repeated requests.