Introducing Hasktorch
summary
Hasktorch is a Haskell library that binds libtorch (the PyTorch C++ backend) to give GPU support and automatic differentiation while exposing both an untyped API and gradually typed tensors for shape-level type safety. It can load TorchScript models or parse state_dicts to fine-tune in Haskell, and deploy via standard web stacks (example uses Servant) with SQLite for storage. Claim: typed functional composition reduces integration fragility; example projects include program synthesis, neperian functors work, and a pre‑alpha self‑hosted knowledge app.
AI-written summary · about
chapters (9)
- 0:00Introduction to Hasktorch
- 1:51Why Haskell for machine learning
- 10:06Neural networks as programs
- 11:51Practical approach to Hasktorch
- 14:57Working with Hasktorch
- 19:31Training models in Hasktorch
- 20:15Building ML applications
- 24:28Hasktorch community and use cases
- 26:57Q&A and resources
related talks
05mWhat's New in the Torch Ecosystem
08mTorchText
43mStorch - GPU Accelerated Deep Learning for Scala 3
47mGet to Know PyTorch - A Cutting Edge Open Source AI Framework for Deep Learning
18mLightning Talk: TorchRL - RLHF Support
19mGet to know PyTorch - a cutting edge open source AI framework for Machine Learning
48mFireside Chat
24mHelion 1.0: A High-Level DSL for Performance Portable Kernels