Disclaimer
This repository was not created using “vibe coding.”
nanoclip is a minimalist, nano-scale implementation of the Contrastive Language-Image Pre-training (CLIP) model, built using PyTorch. This project aims to provide a clear and concise example of how to implement a dual-encoder model that learns to embed images and text into a shared, multimodal latent space.
- Vision Encoder: A lightweight Vision Transformer (ViT) for processing image inputs.
- Text Encoder: A compact Transformer model for encoding text inputs.
- Contrastive Learning: Implements the core CLIP training objective to align image and text embeddings.
- Configurable: Parameters for both vision and text models are easily adjustable via
config.py.
Currently the project is in development, only one toy training run has been done on a remote 1xRTX 3600 on CIFAR10. More to come..
