EmbeddingGemma-300M — unmodified mirror
An ungated mirror of google/embeddinggemma-300m
at revision 57c266a740f537b4dc058e1b0cda161fd15afa75,
published by Looking Glass Labs LLC so that the Busco desktop app can
fetch it on first run without requiring each user to file an individual access request.
Upstream is gated (gated: manual), which means every download needs a per-account approval
from a human. A locally-installed desktop application cannot ask its users to do that, so this
mirror exists to make the same weights retrievable anonymously. It is not a fine-tune, a
quantization, a conversion, or a re-training — there is nothing here that is not upstream.
Modification notice (Gemma Terms of Use §3.1(3))
Every weight, tokenizer, pooling and configuration file in this repository is byte-identical to the upstream revision named above; their Git LFS SHA-256 object ids match upstream's, file for file.
Two files differ, and only these two:
| File | Difference |
|---|---|
README.md | This card replaces the upstream model card. |
notebook.ipynb | Not mirrored. |
Two files are added, and only these two:
| File | Purpose |
|---|---|
LICENSE | The Gemma Terms of Use, reproduced in full (§3.1(2)). |
NOTICE | The notice required by §3.1(4). |
No weight file, tokenizer file or configuration file has been modified, added or removed.
Licence
Gemma is provided under and subject to the Gemma Terms of Use found at
ai.google.dev/gemma/terms. Use of these weights is
additionally subject to the
Gemma Prohibited Use Policy. Both are
incorporated by reference. See LICENSE and NOTICE in this repository for the full text
and the required notice.
Distributing these weights onward carries the same four conditions of §3.1 that this mirror
meets: pass the §3.2 use restrictions through as an enforceable provision of your own
agreement, give your recipients a copy of the Agreement, mark any files you modify, and ship
the NOTICE file.
Use
from sentence_transformers import SentenceTransformer
# Always pin a revision. This repository will not be force-pushed or squashed,
# so any commit sha in its history stays resolvable.
model = SentenceTransformer("busco-app/embeddinggemma-300m", revision="<commit sha>")
For anything other than reproducibility, prefer upstream — it is the canonical source, it carries Google's own model card, evaluation numbers and usage guidance, and none of that is reproduced here.