transformers-convert

# Hugging Face Transformers Model Conversion Convert custom PyTorch models to Hugging Face Transformers format while maintaining exact compatibility with the original implementation. ## Overview This skill provides a systematic workflow for transformers conversion: - Extract hardcoded values into PretrainedConfig - Create PreTrainedModel wrapper - Build ImageProcessor/Tokenizer - Test equivalence thoroughly - Prepare for Hub upload **Important**: Use validation mode (parallel implementations) first to verify equivalence, then replace the original. ## Conversion Workflow ### Step 1: Analyze the Custom Model Ask the user to specify: - Path to the custom model implementation - Model type (vision, text, multimodal) - Task (classification, segmentation, generation, etc.) - Validation mode or replacement mode Then identify: - Model architecture and components - Input/output formats - Key hyperparameters and hardcoded values - Pretrained weights location - Preprocessing pipeline - Custom layers or modules ### Step 2: Create PretrainedConfig Class **Key principle**: Extract ALL hardcoded values from the model as configurable parameters. Template: ```python from transformers import PretrainedConfig from typing import List, Optional class {ModelName}Config(PretrainedConfig): model_type = "{model_name}" def __init__( self, # Core architecture parameters hidden_dim: int = 128, num_layers: int = 4, # Input/output parameters...