Can I still use backend-specific features like CUDA cache clearing?

Yes, the skill includes patterns for backend-specific guards that allow you to call library-specific features (like torch.cuda.empty_cache()) while keeping the main compute path agnostic.

How does this skill help with memory optimization?

It provides specific patterns like detached sampling using stop_gradient to prevent unnecessary backpropagation through sampling paths, which significantly reduces activation memory usage.

How do I handle boundaries between NumPy and Keras tensors?

The skill enforces the use of keras.ops.convert_to_tensor() at the input boundaries and specific casting rules to ensure NumPy arrays are correctly integrated into the Keras compute graph.

What is the benefit of the Straight-Through Estimator (STE) pattern?

The STE pattern allows you to use hard indicators (like thresholding) in your forward pass while maintaining identity gradients in the backward pass, making non-differentiable operations compatible with gradient descent.

Why should I use keras.ops instead of PyTorch or JAX directly?

Using keras.ops ensures your code is backend-agnostic, allowing it to run seamlessly on PyTorch, JAX, or TensorFlow backends without requiring any code modifications.

BayesFlow Keras Ops

Name: BayesFlow Keras Ops
Author: matthiaskloft

bymatthiaskloft

0•

データサイエンスとML

Ensures backend-agnostic tensor mathematics using Keras 3 operations for BayesFlow extensions and custom layers.

The BayesFlow Keras Ops skill provides Claude with the specialized knowledge required to write portable, multi-backend tensor math code within the BayesFlow ecosystem. By enforcing the use of keras.ops over backend-specific libraries like PyTorch, JAX, or NumPy, it ensures that loss functions, custom layers, and network forward passes remain compatible across different computing environments. This skill is essential for developers building differentiable probabilistic models that must remain performant and portable, offering specific implementation patterns for common BayesFlow tasks such as detached sampling, chunked log-probability calculations, and straight-through estimators.

主な機能

01Enforces correct type casting and shape handling using Keras 3 best practices

02Prevents common pitfalls like mixing NumPy and Keras tensors or using backend-specific methods like .view() or .detach()

03Provides implementation patterns for memory-efficient detached sampling and log-prob chunking

040 GitHub stars

05Implements the Straight-Through Estimator (STE) pattern for differentiable hard indicators

06Standardizes tensor math using keras.ops for multi-backend compatibility (JAX, PyTorch, TensorFlow)

ユースケース

01Refactoring existing PyTorch or NumPy research code into Keras 3 backend-agnostic production code

02Writing custom Keras layers or loss functions for BayesFlow that need to run on different compute backends

03Optimizing memory usage during posterior sampling by implementing detached gradient paths and stop_gradient patterns

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthiaskloft/claude-skills keras-ops

For use in Claude.ai and ChatGPT

主な機能

01Enforces correct type casting and shape handling using Keras 3 best practices

02Prevents common pitfalls like mixing NumPy and Keras tensors or using backend-specific methods like .view() or .detach()

03Provides implementation patterns for memory-efficient detached sampling and log-prob chunking

040 GitHub stars

05Implements the Straight-Through Estimator (STE) pattern for differentiable hard indicators

06Standardizes tensor math using keras.ops for multi-backend compatibility (JAX, PyTorch, TensorFlow)

ユースケース

01Refactoring existing PyTorch or NumPy research code into Keras 3 backend-agnostic production code

02Writing custom Keras layers or loss functions for BayesFlow that need to run on different compute backends

03Optimizing memory usage during posterior sampling by implementing detached gradient paths and stop_gradient patterns