# CLIP lecture: sources and reproducible examples

- [CLIP paper — Radford et al., 2021](https://proceedings.mlr.press/v139/radford21a.html)
- [Original OpenAI implementation](https://github.com/openai/CLIP)
- [Linear-probe evaluation](https://github.com/openai/CLIP#linear-probe-evaluation)
- [Image credits](figures/ATTRIBUTION.md)
- [Interactive example: exact vectors, matrices and losses](output/interactive-lecture.json)
- [Saved pretrained measurements](output/embedding-evidence.json)
- [Before/after hat calculation](output/closing-hat-calculation.json)
- [Python code](output/closing-code.py)
- [Simple training notebook](notebooks/clip-simple.ipynb)
- [CLIP loss interactive](https://nipunbatra.github.io/interactives/clip-loss/)
- [Pretrained CLIP playground](https://nipunbatra.github.io/clip-lab/playground/)

The cat, dog and car vectors in the training walkthrough are chosen teaching values. The application demos use saved pretrained measurements, and the before/after hat calculation reuses those same vectors. Both are labelled in the lecture. The complete loss interactive is embedded in the HTML.