Show HN: Charl – ML language with native tensors and autograd

3 points

7 months ago

I built Charl, a compiled language for machine learning with tensors and autograd as native features (not a library). It's alpha-quality but functional.

What works: Neural network training, 22x faster than PyTorch CPU, GPU support, type-safe.

What doesn't: Small ecosystem, alpha bugs, missing features (modules, etc).

Not trying to replace PyTorch - this is an experiment to see if ML-native language design offers advantages.

Try it: https://charlbase.org Code: https://github.com/charlcoding-stack/charlcode

Looking for testers and feedback. What would make this useful for your work?