Just an engineer building cool things
Joined August 2018
- Tweets9
- Following124
- Followers133
- Likes0
Meet Octagon, the infrastructure for your agentic quantitative trader.
Run an agent locally that trades your assets while you work on your laptop.
Free and open source. Connection to trading venues coming soon.
github.com/LucasSte/octagon
It is possible to use MLX as a backend for PyTorch. The performance gains are small, since the bottleneck is not CPU-GPU data transfers. I made the MlxTorch experiment to analyze whether we would get any advantage from the integration.
github.com/LucasSte/MlxTorch
Having a MacBook, I wanted to know which framework, PyTorch or MLX, was faster for AI projects, so I benchmarked them in common tasks. Check the results:
Spoiler: Although PyTorch has more mature tools, MLX is quicker!
github.com/LucasSte/MLX-vs-P…
I wrote a little more about the project on Medium.
medium.com/@lucas.ste/how-to…
After missing the research work I did while I was an undergraduate student, I tried a side project on AI again.
I created a library called Havana that generates a knowledge graph from a LLM attention matrices. Any feedback would be highly appreciated. github.com/LucasSte/havina
After missing the research work I did while I was an undergraduate student, I tried a side project on AI again.
I created a library called Havana that generates a knowledge graph from a LLM attention matrices. Any feedback would be highly appreciated. github.com/LucasSte/havina