Rearchitecting LLMs.
My work is focused on two main pillars: making LLMs more efficient and making them more fair.
| Type | Description |
|---|---|
| 📖 Engineering | Rearchitecting-LLMs: Code on advanced structured pruning and optimization. |
| 🔬 Research | Adaptive Attention Bypass (AAB): Prototype of a dynamic inference system that adjusts attention layers based on input complexity. |
| 📄 Preprint | Exploring GLU Expansion Ratios in Llama-3.2 Models: My published methodology for structured pruning. |
| ✍️ Article | How to Prune LLaMA 3.2 and Similar LLMs: A practical guide on Towards Data Science. |
| Type | Description |
|---|---|
| 🔥 Flagship Library | OptiPfair: My open-source library for fairness-aware model pruning and bias visualization. Try the Interactive Demo on HF Spaces! |
| 🔬 Research | Fairness Pruning Notebook: Implementation of a methodology for targeted neuron pruning to mitigate demographic bias. |
| ✍️ Article | Fairness Pruning: Precision Surgery to Reduce Bias in LLMs: My core methodology published on Towards Data Science. |
|
|
I wrote "Large Language Models Projects" (Apress, 2024) to provide practical, hands-on guidance for AI practitioners.
The accompanying LLM Course on GitHub is a community resource with over 1,700 stars. |
I am currently focused on my research and writing. I am open to select, high-impact collaborations where I can provide strategic value through Technical Advisory or Applied Research Partnerships.
- 🔗 LinkedIn: Pere Martra
- 🤗 Hugging Face: oopere
- ✍️ Medium: @peremartra
- 📧 Direct Inquiries:
peremartra [at] uadla [dot] com