Storch - GPU Accelerated Deep Learning for Scala 3
summary
Storch is a new Scala library that wraps libtorch (via JavaCPP) to provide an idiomatic Scala API, enabling GPU accelerated tensor operations by swapping a dependency and requiring explicit native memory frees for stable GPU runs. The demo trains a ResNet cats-vs-dogs model to ~90% accuracy in about a minute using automatic differentiation and standard optimizers. Tensors are Tensor[A] with compile-time promotion rules implemented using Scala 3 match types, letting many dtype errors surface at compile time.
AI-written summary · about
chapters (7)
- 0:00AI and Storch intro
- 5:30What is Storch
- 7:41Image classification demo
- 12:06Tensors in Storch
- 18:59Training AI models
- 34:18How Storch works
- 40:24Storch future plans
related talks
32mCyfra: Scala on GPU
52mPractical Uses of Scala Tensorflow for Image Analysis
27mIntroducing Hasktorch
22mBringing Deep Learning into the Big Data Analytics
35mTowards Typesafe Deep Learning in Scala
39mDive into Deep Learning with Scala
39mFlare & Lantern: Accelerators for Spark and Deep Learning
04mInterview