| Name: Tensorflow Hub Models |
| Short Name: tfhub_models |
| URL: https://tfhub.dev/ |
| Date: 2023/08/02 |
| License: Apache 2.0 |
| License File: LICENSE |
| Version: 2023/08/02 |
| Security Critical: No |
| Shipped: no |
| CPEPrefix: unknown |
| |
| Description: |
| This directory contains models downloaded from https://tfhub.dev/ under its |
| Apache license. These models are not built into Chromium and are only used as |
| testdata. |
| |
| Uploading New Models: |
| 1. Put all the models you want uploaded (including existing models to keep) in |
| the existing //third_party/tfhub_models/models directory. |
| |
| 2. `cd` to //third_party/tfhub_models. |
| |
| 3. Run the following commands: |
| ``` |
| tar -czf models.tar.gz models/* |
| sha1sum models.tar.gz | awk '{print $1}' > models.tar.gz.sha1 |
| ``` |
| |
| 4. Upload the tarball to the cloud storage bucket: |
| ``` |
| upload_to_google_storage.py --bucket chromium-tfhub-models models.tar.gz |
| ``` |
| |
| 5. Overwrite chromium's sha1 models.tar.gz.sha1 file with the new one. |
| |
| Models: |
| Please keep this list and BUILD.gn up to date when new files are added to the |
| Cloud Storage Bucket. |
| |
| * lite-model_universal-sentence-encoder-qa-ondevice_1.tflite - A TFLite |
| Support compatible embedding model trained on Q&A text, pulled from |
| https://tfhub.dev/google/lite-model/universal-sentence-encoder-qa-ondevice/1. |
| Requires sentencepiece tokenization to be built into Chrome, which is not |
| currently supported. See also https://crrev.com/c/4722355. |
| |
| * rehead_embedding_from_mobilebert.tflite - A TFLite Support compatible |
| embedding model which was adapted from this BERT model, |
| https://tfhub.dev/tensorflow/lite-model/mobilebert/1/metadata/1. The base |
| model was re-headed to expose the embedding layer. |