📘 SBBD 2025 Short Course – Notebooks

This folder contains the teaching notebooks for the course Introduction to LLM-Based Agents (SBBD 2025).

🛠️ Installation

First, create and activate the conda environment, then install dependencies:

pip install -r requirements.txt

## 🖥️ Running on CPU vs. GPU

All HuggingFace models in these notebooks are configured to run on the **CPU**:

```python
model_kwargs={"device": "cpu"}

Why CPU?

• Some entry-level GPUs (e.g., GeForce GT 1030) do not support the CUDA compute capabilities required by recent PyTorch and Transformers builds.
• Using CPU ensures the notebooks run consistently across all machines, including student laptops without dedicated GPUs.
• Results are identical between CPU and GPU. The only difference is speed (GPU can be faster, when supported).

Can I use GPU?

Yes — if your GPU is compatible and you have a working CUDA installation, simply remove the explicit {"device": "cpu"} option. For example:

embedding_model = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2"
)

or for text generation:

generator = pipeline("text-generation", model="gpt2")  # GPU if available

📂 Notebook Index

• 00_intro.ipynb — Environment check and first LLM calls
• 01_ngram_vs_llm.ipynb — classic n-gram models x LLMs
• 02_minimal_agent.ipynb — LLM as brain, agent as body
• 03_prompting_patterns.ipynb — Prompting and interaction patterns
• 04_tool_calling.ipynb — Structured tool calling with LangChain
• 05_rag_pipeline.ipynb — Retrieval-Augmented Generation (RAG) with ChromaDB
• 06_text_to_sql.ipynb — Text-to-SQL pipeline with SQLite