Generative Extraction & Classification using Document Understanding in cross-platform projects - Public Preview

It is now possible to allocate AI Units on a tenant level, details here: Automation Cloud - Allocating licenses to tenants
I believe this should solve your issue (as you can provide access to a certain tenant only to users requiring it/allowed to consume AI Units) - but let me know otherwise :slight_smile:

I am using UiPath Studio 23.4.5 with on-prem Orchestrator license. After installing the DocumentUnderstanding.Activities package , the two activites are not working. Am I missing anything? Really appreciate your help.

Created a community account in automation cloud and using the Web Studio to test this package.
Document Extractor activity is failing even though the PDF file size is only 353 KB. It has 9 pages.

Please refer to the below error message.

2023-12-16 10:38:55.586 GMT−6: Error: UiPath.DocumentUnderstanding.ML.Activities.Exceptions.MLRequestException: The document exceeds the maximum allowed size for a request at UiPath.IntelligentOCR.StudioWeb.Activities.ExtractDocumentDataWithDocumentDat

Really appreciate any help.
Thanks

Hi @Monica_Secelean! Hoping you can help us here.

We have coded a few automations using Gen Classifier and Extractor for v 23.10. The results are great but not that accurate for unstructured invoices etc. but it’s way better than training models using DU. Are you able to share what is the LLM and version of the GPT behind the Classifier and Extractor? I thought I had seen on some search on Google that it may be OpenAI ChatGPT v3. If I manually upload an invoice and ask OpenAI ChatGPT v4 outside of UiPath using the same prompts, I am getting amazing results. Are there plans to upgrade to newer models for better accuracy? Thank you

Hi @Monica_Secelean,

With regard to @Rex_Patel 's question about the LLM would UiPath consider an option where we could access the various model using differing AI Units to account for the prices from Microsoft?

Pricing details:
Language models
Models Context Input (Per 1,000 tokens) Output (Per 1,000 tokens)
GPT-3.5-Turbo-0125 16K 0.0005 0.0015
GPT-3.5-Turbo-Instruct 4K 0.0015 0.0020
GPT-4 8K 0.03 0.06
GPT-4 32K 0.06 0.12

@jgraham we are working on enabling you the use of various LLMs within our products, having them “embeddable” in the DU framework - if it’s any particular use case you’re looking for, do let me know :slight_smile:

@Rex_Patel we keep experimenting with multiple models for increased accuracy - if your facing an issue with the results, would you mind sharing the use case/prompts/docs for us to reproduce and fix it? :slight_smile:

@Monica_Secelean When we try to publish project with Extract Document Data using Generative Extractor, getting below error Invoke Workflow File: Could not load file or assembly ‘CustomGptDocumentTyp.FSHAF3UAlKQ1SEvZe2dcKhP, Version=1.0.0.0, Culture=neutral, PublicKeyToken=null’. The system cannot find the file specified.

The same code works fine from Studio. But when running in attended mode from assistant in same machine , getting above exception.

Hey, we are trying to use the Generative Classifier, we are getting this error when running the automation after the time out: could not retrieve a tenant key

Ai units are available as we are using DU to extract data.

@gowrishankar thanks for your feedback, can you maybe tell us how to reproduce the issue or maybe share the workflow with us?

@Thomas_Mitchell1 thanks for your feedback, can you maybe tell us how to reproduce the issue or maybe share the workflow with us?

@gowrishankar actually our recommendation is for you to check if the error is still reproducible, as we believe we’ve fixed it ever since your comment :slight_smile:

Hi @Monica_Secelean ,

Issue is with studio version 23.4 where dynamic objects are not loaded in the assembly libraries. With latest studio version 23.10 , this is working fine without issues

1 Like

This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.