Become an AI engineer

# Model optimization

Model optimization is a unit of 9 lessons in Rabbithole's Become an AI engineer course, at levels 3 to 5.

## Lessons

1. Hosted fine-tuning: what it is, what it costs, how to steer it · Level 3
2. Model export and formats · Level 3
3. Model Merging: Many Models, One Set of Weights · Level 4
4. Parameter-efficient fine-tuning: train a little, keep the rest · Level 4
5. Post-training quantisation: splitting the difficulty · Level 4
6. Numeric precision formats: bits, ranges and what to compute in · Level 4
7. Preference optimisation in practice · Level 4
8. Pruning and sparsity: removing weights without breaking the model · Level 5
9. Quantisation-aware training · Level 5

## Sources

These lessons draw on 26 sources. The most used:

- [Optimum documentation](https://huggingface.co/docs/optimum/index)
- [Scaling Laws for Precision](https://arxiv.org/abs/2411.04330)
- [PEFT documentation](https://huggingface.co/docs/peft/index)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://arxiv.org/abs/2306.00978)
- [DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence](https://arxiv.org/abs/2606.19348)
- [DeepSpeed — Getting Started and Tutorials](https://www.deepspeed.ai/getting-started/)
- [SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models](https://arxiv.org/abs/2211.10438)
- [SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot](https://arxiv.org/abs/2301.00774)

[All units in Become an AI engineer](https://rabbithole.pro/courses/ai-engineer)

Source page: https://rabbithole.pro/courses/ai-engineer/model-optimization