CLIP -- base (ViT-B/32)
contrastive image-text pretraining, 151M parameters across two towers

Copyright (c) 2021 OpenAI
Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh,
Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark,
Gretchen Krueger, Ilya Sutskever.

Source:  https://openaipublic.azureedge.net/clip/models/40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af/ViT-B-32.pt
Project: https://github.com/openai/CLIP
Paper:   Learning Transferable Visual Models From Natural Language Supervision
         (ICML 2021), arXiv:2103.00020
Licence: MIT (full text in the LICENSE file beside this one)

The repository's LICENSE is MIT and no separate weights licence is published, so the
checkpoints are covered by it. That is an inference from silence, recorded as one.

This file is not the byte stream OpenAI serves. Upstream publishes a TorchScript
archive; tools/fetch/clip.py verifies it against the sha256 in OpenAI's own URL, then
recovers the state dict with torch.jit.load and writes the tensors back out as an
ordinary checkpoint, cast to fp32. No tensor is otherwise altered. The archive's sha256
is 40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af.