# Upstream license declarations and provenance

This distribution contains a fine-tuned derivative of
[`vllm-sr/Vela-1.0-Encoder-307M`](https://huggingface.co/vllm-sr/Vela-1.0-Encoder-307M),
using revision `720ab37904ce15054068d429381bfae7549f00f6`.
That revision's model card declares `license: mit`. Its upstream model is
[`jhu-clsp/mmBERT-base`](https://huggingface.co/jhu-clsp/mmBERT-base), whose queried
model card also declares MIT.

The queried pinned Vela files and upstream mmBERT files provide no separate
LICENSE or NOTICE file. This omission is recorded rather than manufacturing
an upstream copyright notice or assigning ownership of their weights to the
fine-tuning author. The model license file contains the standard MIT permission
and warranty terms, sourced from the
[SPDX MIT license text](https://github.com/spdx/license-list-data/blob/main/text/MIT.txt).
Any authoritative upstream copyright notices subsequently supplied should be
preserved with this distribution. These statements reflect published license
declarations rather than a comprehensive audit of upstream training corpora.

The new classifier head and final two encoder layers were updated in a personal
training experiment. This derivative is not an official upstream release and
does not imply endorsement by vLLM Semantic Router or JHU.

The accompanying reproduction source and synthetic dataset retain Apache-2.0
licensing from the Semantic Router experiment's source/data context. Their
license copies are separate from the MIT model license. No private reference
document, original test examples or credentials are distributed.