📘 Learn Mistral — Practical AI Workshops

Welcome to the official repository for Learn Mistral book published by Packt, a hands-on, workshop-driven guide to mastering open-source large language models (LLMs). This repo supports the book's chapters (2–11) and contains code, datasets, configs, and deployment files for practical experimentation.

📍 Repo URL: https://github.com/PacktPublishing/Learn-Mistral

---

📂 Repository Structure

Each chapter has its own subfolder (e.g., Chapter-3, Chapter-5) and includes self-contained workshop materials. Below is a breakdown by chapter:

---

Chapter-2 – Model Setup Examples

• Folders: 7b, mistral-nemo, mistral-small
• Each contains: emoji.modelfile — Sample model definitions or tokenizer configs for Mistral variants.

---

Chapter-3 – Softmax and Visualization

• ch03_plot_pie_chart.py: Visualize categorical distributions.
• ch03_softmax_workshop.py: Simulates softmax behavior with step-by-step logic.

---

Chapter-4 – Embeddings in Action

• ch04_embeddings.ipynb: Compute and compare vector embeddings using popular libraries.

---

Chapter-5 – Fine-Tuning and Evaluation

• ch05_ws1_*.py/.ipynb: Loan prediction pipeline with and without Weights & Biases (W&B).
• ch05_ws1_loaners-repayment-data.csv: Dataset for workshop 1.
• ch05_ws2_mistral_finetune_7b.ipynb: Fine-tune Mistral 7B.
• ch05-mistarl-7b-train-config.yaml, *-validation-config.yaml: Configuration files for model training & validation.

---

Chapter-6 – Retrieval-Augmented Generation (RAG)

All notebooks use either ChromaDB or Pinecone for retrieval backend:

• Ch06_WS1: Basic RAG
• Ch06_WS2: Multi-query RAG
• Ch06_WS3: Chain-of-thought and decomposition
• Ch06_WS4: Step-back prompting and generalization

---

Chapter-7 – Codestral App Deployment

• main.py: Web backend using Codestral models.
• config.json: Application config.
• templates/joke.html: Frontend template for joke generation.

---

Chapter-8 – Secure & Vulnerable Code Patterns

• JdbcServlet.java: Java with unsafe JDBC.
• too-permissive.terraform: Example of misconfigured infrastructure as code.
• vulnerable.php: PHP with common security flaws.

---

Chapter-10 – Cloud LLM Invocation

• requirements.txt: Required Python packages.
• ws10_step01_list_models.py: Lists cloud-available models.
• ws10_step02_call_model.py: Calls/invokes selected models via API.

---

Chapter-11 – Dockerized Local LLM Deployment

• docker-compose-*.yml: Run models like Mistral 7B, 8B, and Bloom locally via Docker.
• ch11_ws_step3_*.py: Scripts to test and interact with deployed models.

---

🚀 How to Use

git clone https://github.com/PacktPublishing/Learn-Mistral.git
cd Learn-Mistral/Chapter-5
python ch05_ws1_loaners_predict.py

Or open .ipynb notebooks directly in JupyterLab or VSCode.

---

🧰 Requirements

Dependencies vary by chapter. Most workshops use:

• transformers, torch, faiss, langchain, pinecone-client
• For fine-tuning: wandb, peft
• Deployment: docker, uvicorn, fastapi

Refer to requirements.txt (where present) or the first cell of each notebook.

---

📬 Contributing

Pull requests and issues are welcome. If you have improvements, bug reports, or suggestions for new workshops — open an issue or PR!