martynassubo

5 ms·

1. 1.▲
   How Trillion Parameter Models Are Trained(martynassubonis.substack.com)

   3 points by martynassubo 2y ago|0 comments
2. 2.▲
   Distributed Data Parallel Training(martynassubonis.substack.com)

   1 points by martynassubo 2y ago|0 comments
3. 3.▲
   by martynassubo 2y ago

   While uv wasn't very mature, I wrote a similar piece, where in the end I tried suggesting pyenv/pipx/poetry: https://open.substack.com/pub/martynassubonis/p/python-proje... However, these days,
4. 4.▲
   Lightweight Model Serving(github.com)

   1 points by martynassubo 2y ago|1 comments
5. 5.▲
   by martynassubo 2y ago

   Hello, I recently completed a small project where I explored the potential improvements in model serving by switching from naive PyTorch setups to the ones utilizing ONNX Runtime, as well as ONNX Runtime with Rust's Actix web server. H
6. 6.▲
   by martynassubo 3y ago

   [1] The failure of pysandbox: https://lwn.net/Articles/574215/
7. 7.▲
   Ask HN: Secure Python code execution environment

   1 points by martynassubo 3y ago|1 comments