72 lines
2.8 KiB
Markdown
72 lines
2.8 KiB
Markdown
# LLMs for Engineers
|
||
|
||
**CHEG 667-013 — Chemical Engineering with Computers**
|
||
Department of Chemical and Biomolecular Engineering, University of Delaware
|
||
|
||
A hands-on workshop on Large Language Models and machine learning for engineers. Learn how to train a GPT from scratch, run local models, build retrieval-augmented generation systems, connect LLMs to tools to build agentic systems, and explore the underlying machine learning methods by implementing a simple neural network.
|
||
|
||
## Sections
|
||
|
||
| # | Topic | Description |
|
||
|---|-------|-------------|
|
||
| [01](01-nanogpt/) | **nanoGPT** | Train a small transformer on Shakespeare. Explore model parameters, temperature, and text generation. |
|
||
| [02](02-ollama/) | **Local models with Ollama** | Run pre-trained LLMs locally. Summarize documents, query arXiv, generate code, build custom models. |
|
||
| [03](03-rag/) | **Retrieval-Augmented Generation** | Build a RAG system: chunk documents, embed them, and query with an LLM grounded in your own data. |
|
||
| [04](04-semantic-search/) | **Advanced retrieval** | Hybrid BM25 + vector search with cross-encoder re-ranking. Compares summarization versus raw retrieval. |
|
||
| [05](05-tool-use/) | **Tool use and agentic systems** | Connect an LLM to Python functions via tool calling. Build a thermo assistant. Understand how ChatGPT, Claude, and Copilot actually work. |
|
||
| [06](06-neural-networks/) | **Building a neural network** | Implement a one-hidden-layer network from scratch in numpy, then in PyTorch. Fits $C_p(T)$ data for N₂. |
|
||
|
||
|
||
|
||
## Prerequisites
|
||
|
||
- A terminal (macOS/Linux, or WSL on Windows)
|
||
- Python 3.10+
|
||
- Basic comfort with the command line
|
||
- [Ollama](https://ollama.com) (sections 02–05)
|
||
|
||
## Getting started
|
||
|
||
Clone this repository and work through each section in order:
|
||
|
||
```bash
|
||
git clone https://lem.che.udel.edu/git/furst/llm-workshop.git
|
||
cd llm-workshop
|
||
```
|
||
|
||
Each section has its own `README.md` with a full walkthrough, exercises, and any code or data needed.
|
||
|
||
### Python environment
|
||
|
||
Install [uv](https://docs.astral.sh/uv/getting-started/installation/) (a fast Python package manager), then:
|
||
|
||
```bash
|
||
uv sync
|
||
```
|
||
|
||
This creates a `.venv/` virtual environment and installs all dependencies from the lock file.
|
||
|
||
**Note:** On Apple Silicon Macs, PyTorch GPU acceleration (MPS) works out of the box. On NVIDIA GPU machines, the default `uv sync` install may be CPU-only and you need to reinstall with CUDA support. See [PYTORCH.md](01-nanogpt/PYTORCH.md) for troubleshooting and device-specific instructions.
|
||
|
||
`cd` into the section directory before running scripts or notebooks, since they reference local data files:
|
||
|
||
```bash
|
||
cd 06-neural-networks
|
||
uv run python nn_torch.py
|
||
```
|
||
|
||
Or activate the environment and run directly:
|
||
|
||
```bash
|
||
source .venv/bin/activate
|
||
cd 06-neural-networks
|
||
python nn_torch.py
|
||
```
|
||
|
||
## License
|
||
|
||
MIT
|
||
|
||
## Author
|
||
|
||
Eric M. Furst, University of Delaware
|