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

Yes, the skill provides specific code guards to safely implement backend-specific logic (like torch.cuda.empty_cache) without breaking compatibility for other backends.

Does this skill handle gradient stopping?

Yes, it includes specific patterns for using stop_gradient to save memory during sampling and for implementing Straight-Through Estimators (STE).

How do I convert a NumPy array to a Keras tensor using this skill?

The skill guides you to use keras.ops.convert_to_tensor at the boundary of your compute paths to ensure seamless data integration.

What common tensor mistakes does this skill prevent?

It identifies and corrects the use of framework-specific methods such as .view(), .item(), .float(), or using Python's sum() on Keras tensors.

Why should I use keras.ops instead of torch or numpy?

Using keras.ops ensures your code is portable and can run on PyTorch, JAX, and TensorFlow backends without modifications, which is critical for the BayesFlow ecosystem.

Keras Backend-Agnostic Math

Name: Keras Backend-Agnostic Math
Author: matthiaskloft

bymatthiaskloft

0•

数据科学与机器学习

Standardizes tensor math across PyTorch, JAX, and TensorFlow backends for BayesFlow extensions using Keras 3 operations.

This skill ensures that all tensor-level computations in BayesFlow-related code remain backend-agnostic by strictly enforcing the use of Keras 3 operations over framework-specific libraries like PyTorch or JAX. It provides a comprehensive mapping of common math functions, implementation patterns for detached sampling, and strategies for handling memory-intensive log-probability calculations, helping developers maintain portable and robust deep learning models within the BayesFlow ecosystem.

主要功能

01Manages safe boundary crossings between NumPy arrays and Keras tensors

02Prevents common framework-specific mistakes such as using .view(), .item(), or torch.cat()

03Implements advanced patterns like Straight-Through Estimators (STE) and chunked log-probabilities

040 GitHub stars

05Enforces keras.ops for universal compatibility across PyTorch, JAX, and TensorFlow

06Provides a quick-reference mapping for common tensor operations like sum, mean, and reshape

使用场景

01Developing custom Keras layers or loss functions for BayesFlow that must support multiple backends

02Optimizing memory usage in inference networks using gradient-detached sampling patterns

03Refactoring existing PyTorch or NumPy math code into backend-agnostic Keras 3 logic

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

主要功能

01Manages safe boundary crossings between NumPy arrays and Keras tensors

02Prevents common framework-specific mistakes such as using .view(), .item(), or torch.cat()

03Implements advanced patterns like Straight-Through Estimators (STE) and chunked log-probabilities

040 GitHub stars

05Enforces keras.ops for universal compatibility across PyTorch, JAX, and TensorFlow

06Provides a quick-reference mapping for common tensor operations like sum, mean, and reshape

使用场景

01Developing custom Keras layers or loss functions for BayesFlow that must support multiple backends

02Optimizing memory usage in inference networks using gradient-detached sampling patterns

03Refactoring existing PyTorch or NumPy math code into backend-agnostic Keras 3 logic