Over the last year, Annotation Lab has grown to be much more than a document annotation tool. It became a full-fledged AI system, capable of testing pre-trained models and rules, applying them to new datasets, training, and tuning models, and exporting them to be deployed in production. All those features together with the new Playground concept presented in the current release notes contributed to the transformation of the Annotation Lab into the NLP Lab. A new Playground feature is released as part of the NLP Lab’s Hub that allows users to quickly test any model and/or rule on a snippet of text without the need to create a project and import tasks. NLP Lab also supports the training of Legal and Finance models and Model evaluation for classification projects. As always, the release includes some stabilization and bug fixes for issues reported by our user community. Below are the details of what has been included in this release.
The Playground: Test, Demo, and Serve Models and Rules in NLP Lab
NLP Lab introduces the Playground feature where users can directly deploy and test models and/or rules. In previous versions, the pre-annotation servers could only be deployed from within a given project. With the addition of the Playground, models can easily be deployed and tested on a sample text without going through the project setup wizard. Any model or rule can now be selected and deployed for testing by clicking on the “Open in Playground” button.
For each model, the Playground shows the benchmarking information where available. Benchmarking information are available for pretrained Heathcare, Finance and Legal models as well as for models trained within the NLP Lab.
Rules are deployable in the Playground from the Rules page. When a particular rule is deployed to the Playground, the user can also change the definition of the rules via the form available on the right side of the page. After saving the changes, users need to click on the “Deploy” button to refresh the results of the pre-annotation on the provided text.
Deployment of models and rules is supported by floating and air-gapped licenses. Healthcare, Legal, and Finance models require a license with their respective scopes to be deployed in Playground. Unlike pre-annotation servers, only one Playground instance can be deployed at any given time.
Training and Preannotation with Finance and Legal Models
With this release, users can perform training of Legal and Finance models depending on the available license(s). When training a new model in the NLP Lab, users have the option to select what library to use. Two options were available up until now: Open source and Healthcare. This release adds two new options: Legal and Finance. This helps differentiate the library used for training the models. The new options are only available when at least one valid license with the corresponding scope is added to the License page.
Publish Trained Models to S3
The NLP Lab also allows users to easily export trained models to a given s3 bucket. This feature is available on the Models page under the Hub tab. Users need to enter the s3 bucket path, s3 access key, and s3 secret key to upload the model to the s3 bucket.
Getting Started is Easy
The NLP Lab is a free tool that can be deployed in a couple of clicks on the AWS and Azure Marketplaces, or installed on-premise with a one-line Kubernetes script. Get started here: https://nlp.johnsnowlabs.com/docs/en/alab/install