Birbla

Comments

Hacker News

The idea makes sense because researchers already think in PyTorch. The tooling should adapt to them, not the other way around.

by madhu_ghalame

Anyone not doing distributed training across thousands of GPU, pytorch is the best option. Others have an infra team to work on it.

by insumanth

This kind of sounds like just giving up on making the compiler better.

by ansk

I think this idea is exactly right. Many ML papers that include pseudocode basically just write out a minimal PyTorch training loop - down to the loss.backward(). It has, along with Python, become something of a linga franca for research in the field.

by ainch

Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

Sign in
  • Hacker News
  • The idea makes sense because researchers already think in PyTorch. The tooling should adapt to them, not the other way around.
    by madhu_ghalame
  • this is just https://github.com/tinygrad/tinygrad
    by ofou
  • Anyone not doing distributed training across thousands of GPU, pytorch is the best option. Others have an infra team to work on it.
    by insumanth
  • This kind of sounds like just giving up on making the compiler better.
    by ansk
  • I think this idea is exactly right. Many ML papers that include pseudocode basically just write out a minimal PyTorch training loop - down to the loss.backward(). It has, along with Python, become something of a linga franca for research in the field.
    by ainch