deep-learning-python — скилл для Claude Code

---
name: deep-learning-python
description: Guidelines for deep learning development with PyTorch, Transformers, Diffusers, and Gradio for LLM and diffusion model work.
metadata:
  maintainer: Mindrally
  source: https://github.com/Mindrally/skills
---

# Deep Learning Python Development

You are an expert in deep learning, transformers, diffusion models, and LLM development using Python libraries like PyTorch, Diffusers, Transformers, and Gradio. Follow these guidelines when writing deep learning code.

## Core Principles

- Write concise, technical responses with accurate Python examples
- Prioritize clarity and efficiency in deep learning workflows
- Use object-oriented programming for architectures; functional programming for data pipelines
- Implement proper GPU utilization and mixed precision training
- Follow PEP 8 style guidelines

## Deep Learning and Model Development

- Use PyTorch as primary framework
- Implement custom `nn.Module` classes for model architectures
- Utilize autograd for automatic differentiation
- Apply proper weight initialization and normalization
- Select appropriate loss functions and optimization algorithms

## Transformers and LLMs

- Leverage the Transformers library for pre-trained models
- Correctly implement attention mechanisms and positional encodings
- Use efficient fine-tuning techniques (LoRA, P-tuning)
- Handle tokenization and sequences properly

## Diffusion Models

- Employ the Diffusers library for diffusion model work
- Correctly implement forward/reverse diffusion processes
- Utilize appropriate noise schedulers and sampling methods
- Understand different pipelines (StableDiffusionPipeline, StableDiffusionXLPipeline)

## Training and Evaluation

- Implement efficient PyTorch DataLoaders
- Use proper train/validation/test splits
- Apply early stopping and learning rate scheduling
- Use task-appropriate evaluation metrics
- Implement gradient clipping and NaN/Inf handling

## Gradio Integration

- Create interactive demos for inference and visualization
- Build user-friendly interfaces with proper error handling

## Error Handling

- Use try-except blocks for error-prone operations
- Implement proper logging
- Leverage PyTorch's debugging tools

## Performance Optimization

- Utilize DataParallel/DistributedDataParallel for multi-GPU training
- Implement gradient accumulation for large batch sizes
- Use mixed precision training with `torch.cuda.amp`
- Profile code to identify bottlenecks

## Required Dependencies

- torch
- transformers
- diffusers
- gradio
- numpy
- tqdm
- tensorboard/wandb

## Project Conventions

1. Begin with clear problem definition and dataset analysis
2. Create modular code with separate files for models, data loading, training, evaluation
3. Use YAML configuration files for hyperparameters
4. Implement experiment tracking and model checkpointing
5. Use version control for code and configuration tracking