GPU-Powered ML, AI, and Quant Finance with Google Colab (3)

AI Engineering Workflows with Google Colab — Session 3 🚀 Train a tiny GPT from scratch, fine-tune an LLM with LoRA, build a RAG pipeline, and serve a coding model on the CLI — all on a free Google Colab T4 GPU. In this final session of our free 3-part webinar series, Dr. Yves J. Hilpisch walks through four ready-to-run notebooks that cover the full AI engineering lifecycle: build, adapt, retrieve, and serve. 📘 Notebooks in this session: 09 — yoctoGPT: train a minimal character-level GPT from scratch 10 — LoRA Fine-Tuning: adapt a 360M LLM with only 0.23% trainable parameters 11 — RAG with FAISS: reduce hallucination via retrieval-augmented generation 12 — Coding LLM CLI: serve qwen2.5-coder:7b with Ollama, expose via cloudflared, and use with aider ⚡ Key highlights on a free T4 GPU: • yoctoGPT: train a character-level model on Thoreau's Walden • LoRA: fine-tune with 0.23% of total parameters • RAG: grounded answers with FAISS-backed retrieval • Coding LLM: OpenAI-compatible API from a Colab notebook — and in the live demo, the local aider-chat instance actually wrote a bsm.py file to disk (albeit to the parent directory), proving the full loop works 🔗 Resources: 📑 Slides: https://tpq.io/tpq_colab_3.pdf 📂 Notebook repo: https://github.com/yhilpisch/colab 📈 CPF: https://cpf.tpq.io 🤖 TAE: https://theaiengineer.dev ₿ TCE: https://thecryptoengineer.dev 📊 TDS: https://thedatascientist.dev 📌 Disclaimer: This content is provided solely for illustration and educational purposes. The Python Quants GmbH is not affiliated with, endorsed by, or sponsored by Google. All numbers and specifications are indicative; always check the official Google Colab documentation for current and reliable information. #GoogleColab #GPU #AIEngineering #LoRA #RAG #FAISS #LLM #Ollama #Python #GenerativeAI