ML Studio dependencies
ML Studio relies on training scripts, in the form of an ibformers package, and base models. Both of these dependencies are provided through the Marketplace and must be updated periodically.
-
For new Instabase installations, ML Studio dependencies are typically installed through automated post-install actions, and no further action is required.
-
For upgrades to version 23.01 or later, you must update Marketplace to make available the latest
ibformersfiles and base models.
Dependency compatibility
Instabase versions must be used with a suitable ibformers version.
| Instabase version | ibformers version |
|---|---|
| 24.04 or later | 2.3.2 or later |
| 23.07–24.01 | 2.2.0 or later |
| 22.01–23.04 | Optionally upgrade to 2.2.0, but some features might not work as expected. |
Running Instabase 23.01 or later with an incompatible ibformers package or base models can result in a Model not found in cache error when you attempt to train a model. To resolve this error, update Marketplace.
Upgrade published models
You can upgrade your previously trained models to use the latest version of .ibformers, providing improved inference and security. Upgrades are available for models that use .ibformers 2.0 or later and were trained in and published from ML Studio.
To complete this task, you must be a site administrator or have the Marketplace Admin app permission.
-
In Marketplace Admin, navigate to the model you want to upgrade.
-
Click Upgrade. The
.ibformerspackage is upgraded and the model republished in the Marketplace.
Be sure to test your model and any flow that uses the model after upgrade.
ibformers documentation
The ibformers documentation details components and functionality provided in the ibformers package, including base models, hyperparameters, and metrics. Components and functionality vary by version, so consult the documentation for your ibformers version.
This version of Instabase is compatible with the following ibformers versions:
For details about earlier ibformers versions, see ibformers 2.2.0 or the package documentation in Developer Exchange (deprecated).
Base models
These are the current base models available from Instabase.
Base model names and paths were changed to support updated architecture in which ibformers v2 performs base model lookup and download.
| Current name | Legacy name |
|---|---|
| bert_base_uncased-0.0.1.ibsolution | bert_base, natural_language (merged) |
| instalm_base_draft-0.0.1.ibsolution | instalm_base |
| instalmv3-0.0.3.ibsolution | instalmv3 |
| laymqav1_base-0.0.1.ibsolution | layoutMultiQA_base |
| layoutlm_base_uncased-0.0.1.ibsolution | layout_base, layout_split_classifier (merged) |
| layoutlmv2_base_uncased-0.0.1.ibsolution | layoutv2 |
| layoutlmv3_base-0.0.1.ibsolution | layoutv3 |
| layoutxlm_base-0.0.1.ibsolution | layout_multilingual |
| legal_bert_base_uncased-0.0.1.ibsolution | legal_natural_language |
| table_model_draft_v2-0.0.3.ibsolution | table_detr_base |
Update Marketplace
Updating Marketplace publishes the latest version of ibformers and base models to your Marketplace instance and makes them available for use in your Instabase environment.
To complete this task, you must be a site administrator or have both the Marketplace Admin app permission and the Manage Site Configuration site permission.
-
From the Marketplace Admin app, select Tools > Update Marketplace.
-
In the Docker Image URL field, enter the URL
gcr.io/instabase-public/instabase-artifacts:latestand click Upload From Image. -
Verify that
ibformersand all current base models are present in the directory<instabase_url>/system/global/fs/Instabase Drive/Marketplace/Update/Pending/. -
When all packages are uploaded, return to the Update Marketplace window and click Publish Solutions.
TipIf publishing fails, you can manually upload and publish solution files.
-
From the Marketplace Admin home screen, confirm that
ibformerspackages show version2.x.xand that all current base models are listed.TipNote the latest version listed for
ibformers, because you must specify it in configuration options for model training scripts. -
From the Admin app, click Configuration.
-
Under Model training scripts, set the
"mp_version"key for both extraction and classification to the version ofibformersyou just published, then click Save.