clip_deploy
Copyright (c) 2021 OpenAI

This product includes software developed by OpenAI, published as CLIP
(https://github.com/openai/CLIP) and described in "Learning Transferable Visual
Models From Natural Language Supervision" (Radford, Kim, Hallacy, Ramesh, Goh,
Agarwal, Sastry, Askell, Mishkin, Clark, Krueger, Sutskever; ICML 2021,
arXiv:2103.00020).

The modules here are extracted from that repository, which is licensed under the
MIT License; see the LICENSE file in this directory, which is byte-identical to
the one published there.

ON THE WEIGHTS
--------------
The published checkpoints are MIT. The repository ships one LICENSE and no
separate weights licence, so the checkpoints are covered by it -- an inference
from silence, and recorded as one here rather than presented as a statement the
authors made. Nothing on openaipublic.azureedge.net, where the checkpoints are
served, states terms of its own.

THE VOCABULARY
--------------
assets/bpe_simple_vocab_16e6.txt.gz is OpenAI's byte-pair vocabulary, published
with CLIP and unmodified (sha256 924691ac288e54409236115652ad4aa250f48203de50a9e
4722a6ecd48d6804a, 1,356,917 bytes). It is byte-identical to the copies carried
by mozo.vendors.owlv2_deploy and mozo.vendors.sam3_deploy: a vendored package may
not import another, so this is a third copy of one file rather than a shared
asset.

NOT INCLUDED
------------
The five ResNet image towers (RN50, RN101, RN50x4, RN50x16, RN50x64), the
contrastive training loss, the TorchScript execution path, and the tokenizer's
silent truncation. See PROVENANCE.md for the full list and for why.
