Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 2 Next »

You have decided to leverage the power of LLMs and now are ready to experiment with prompts to get the highest quality results. It is very simple to start with Contract.Fit.

Step 1: Find the Format(s) related to the Project you are interested in fine-tuning

Go to “Studio” → “Projects” → “Edit project document types” and recognize the format names.
Then, go to “Studio” → “Formats” and find the formats of interest.

If you have multiple formats with the same name, and thus you are unsure which one is related to your project, please look for the ID of the format(s) through the API in the “Get project attributes” endpoint. Then, you can make sure the format is the same by checking the JSON output of the format in the format page.

Step 2: Edit the Format(s)

Once you have format(s) of interest, click on “Edit”. There you will see all the tables and fields that are extracted for that document type.

In order to add prompts, you will first have check the “prompt” boxes as in the picture below. You can check only one, multiple, or all of them as needed. When a field is not checked on, it will use the old model for the extractions. At least one field should have the box checked in order for the table prompt to work.

Once the box(es) are checked, you can add a table prompt (in the example above, this would be where “Write here a prompt for the ‘line_items’ table” in written) or/and field prompts (where “Prompt” is written). Once you are happy with the configuration, you can go ahead and click of “Publish Changes”.

Step 3: See the Prompting results

With the format changes successfully published, you can now go to any of the inboxes of your project and either upload or reprocess file(s) to see the results of your prompting.

  1. Upload file(s): Just upload one of many files in the way you prefer, and see the results of your prompting in the default version.

  2. Reprocess file(s): If you want to see the results of your prompting in just one file, you can open the specific file, click on “Actions” and then on “Reprocess”. If you want to reprocess one, multiple or all files in your inbox, you can go to the inbox section, select the desired files with the check boxes at the right, click on “File actions” and “Reprocess”.

Once you click on “reprocess”, you can save the version with the name you prefer, this will allow you to compare results with other versions as you experiment with different prompts. You can also type a version name that already exists if you wish to overwrite it.

Lastly, we do not offer a way to store prompts together with the version at the moment. This means that while the version name and its predictions will be visible to compare, the prompts related to the version are not saved. We recommend you save the prompts that you liked (those that yielded the best results, for example) together with the version name in a separate place so you can go back to them when needed.

  • No labels