Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

InstructLab and Deepsearch #106

Open
wants to merge 1 commit into
base: main
Choose a base branch
from

Conversation

jjasghar
Copy link
Member

This is the proposal to start integrating th document conversion system Deepsearch from IBM Research and InstructLab

@jjasghar jjasghar force-pushed the jjasghar/deepsearch branch 5 times, most recently from 2787217 to ba8aaed Compare June 25, 2024 21:29
This is the proposal to start integrating th document conversion
system Deepsearch from IBM Research and InstructLab

Co-authored-by: Ming Zhao <[email protected]>
Co-authored-by: BJ Hargrave <[email protected]>
Signed-off-by: JJ Asghar <[email protected]>
@jjasghar jjasghar force-pushed the jjasghar/deepsearch branch from ba8aaed to c2ec3e1 Compare June 25, 2024 21:31
Copy link
Member

@hickeyma hickeyma left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @jjasghar for pushing the proposal. Suggestions added inline.


# DeepSearch + InstructLab Integration Proposal

<https://github.com/DS4SD>
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Moved below where DeepSearch is first mentioned

Comment on lines +8 to +12
Managing submissions for the open-source InstructLab project has revealed a significant bottleneck in processing
knowledge documents. For the InstructLab backend to effectively utilize these documents, they must be in markdown
format. Currently, we only accept Wikipedia articles, but the built-in conversion tool is inadequate. Internally at
IBM, and other companies, many knowledge submissions are in multiple document formats, including PDF format,
necessitating conversion to markdown before being used in InstructLab.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
Managing submissions for the open-source InstructLab project has revealed a significant bottleneck in processing
knowledge documents. For the InstructLab backend to effectively utilize these documents, they must be in markdown
format. Currently, we only accept Wikipedia articles, but the built-in conversion tool is inadequate. Internally at
IBM, and other companies, many knowledge submissions are in multiple document formats, including PDF format,
necessitating conversion to markdown before being used in InstructLab.
Managing taxonomy submissions in InstructLab has revealed an issue in processing knowledge documents. For the InstructLab to handle these documents, they must be in [markdown](https://en.wikipedia.org/wiki/Markdown) format. However, many knowledge submissions are in multiple document formats, including PDF, which necessitating conversion to markdown before being used by InstructLab.

Comment on lines +14 to +16
Existing open-source methods, such as PanDoc, are inconsistent. While they preserve text, they struggle with parsing
tables and special symbols, as evidenced by issues in PR #1154 of the taxonomy repo in the InstructLab project. Other
open-source solutions have similar shortcomings.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
Existing open-source methods, such as PanDoc, are inconsistent. While they preserve text, they struggle with parsing
tables and special symbols, as evidenced by issues in PR #1154 of the taxonomy repo in the InstructLab project. Other
open-source solutions have similar shortcomings.
Existing open source tools, such as [Pandoc](https://pandoc.org/), are inconsistent. While they preserve text, they struggle with parsing tables and special symbols, as evidenced by issues in [PR #1154](https://github.com/instructlab/taxonomy/pull/1154) of the taxonomy repo in the InstructLab. Other open-source solutions have similar shortcomings.

Comment on lines +20 to +23
IBM's DeepSearch software excels in document conversion, outperforming traditional open-source methods. Utilizing a
computer vision model layer, it accurately parses content in the files, including titles, headers, and tables.
Additionally, it automatically implements RAG layers for models, which could benefit the InstructLab process in
the future.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
IBM's DeepSearch software excels in document conversion, outperforming traditional open-source methods. Utilizing a
computer vision model layer, it accurately parses content in the files, including titles, headers, and tables.
Additionally, it automatically implements RAG layers for models, which could benefit the InstructLab process in
the future.
IBM's [DeepSearch](https://github.com/DS4SD) software excels in document conversion, outperforming traditional open source methods. Utilizing a computer vision model layer, it accurately parses content in the files, including titles, headers, and tables. Additionally, it automatically implements RAG layers for models, which could benefit the InstructLab process in the future.

@@ -0,0 +1,52 @@

# DeepSearch + InstructLab Integration Proposal
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
# DeepSearch + InstructLab Integration Proposal
# Document Conversion Proposal

Comment on lines +30 to +33
### Open-Source Conversion

- Implement a basic document conversion tool in the UI using an open-source method such as PanDoc. This tool will be
lightweight and easily hosted, ensuring it can be used and improved by the community.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
### Open-Source Conversion
- Implement a basic document conversion tool in the UI using an open-source method such as PanDoc. This tool will be
lightweight and easily hosted, ensuring it can be used and improved by the community.
- Open source conversion: Implement a basic document conversion tool in the InstructLab UI using an open source tool such as [Pandoc](https://pandoc.org/). This tool will be lightweight and easily hosted, ensuring it can be used and improved by the community.

Comment on lines +35 to +39
### DeepSearch Integration

- Enable the UI to switch the conversion endpoint to DeepSearch, allowing high-fidelity markdown conversions for
backend use. This approach maintains an open-source version while benefiting from DeepSearch's superior
conversion capabilities.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
### DeepSearch Integration
- Enable the UI to switch the conversion endpoint to DeepSearch, allowing high-fidelity markdown conversions for
backend use. This approach maintains an open-source version while benefiting from DeepSearch's superior
conversion capabilities.
- [DeepSearch](https://github.com/DS4SD) conversion: Enable the InstructLab UI to switch the conversion endpoint to DeepSearch, allowing high-fidelity markdown conversions for backend use. This approach uses the open source version of Deepsearch while benefiting from DeepSearch's superior conversion capabilities.

Comment on lines +41 to +43
IBM Research and the DeepSearch team will host the DeepSearch endpoint for the open-source community. This
arrangement benefits the community by streamlining contributions and provides data and exposure for the DeepSearch
project. IBM's contribution underscores its commitment to supporting and improving open-source projects.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
IBM Research and the DeepSearch team will host the DeepSearch endpoint for the open-source community. This
arrangement benefits the community by streamlining contributions and provides data and exposure for the DeepSearch
project. IBM's contribution underscores its commitment to supporting and improving open-source projects.
IBM Research and the DeepSearch team will host the DeepSearch endpoint for the InstructLab community. This
arrangement benefits the community by providing a means to handle different document formats.

Comment on lines +45 to +48
This integration will highlight the value of DeepSearch, highlighting their potential for those integrating
InstructLab into their workflows. If the volume of community requests becomes unsustainable for the DeepSearch team,
we hope for ample notification to allow the community to find alternative solutions. By then, we anticipate that the
open-source versions will have improved sufficiently, or the value of the integration will justify continued support.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
This integration will highlight the value of DeepSearch, highlighting their potential for those integrating
InstructLab into their workflows. If the volume of community requests becomes unsustainable for the DeepSearch team,
we hope for ample notification to allow the community to find alternative solutions. By then, we anticipate that the
open-source versions will have improved sufficiently, or the value of the integration will justify continued support.
If the volume of community requests becomes unsustainable for the DeepSearch team to manage, we hope for ample notification to allow the community to find alternative solutions. By then, we anticipate that the open source conversion tools will have improved sufficiently, or the value of the integration will justify continued support.

Comment on lines +50 to +52
By adopting this two-pronged approach, we ensure the integrity of the open-source project while leveraging IBM's
advanced DeepSearch capabilities. This strategy balances community collaboration with innovative technology,
fostering innovation and improvement in document processing for the InstructLab project.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
By adopting this two-pronged approach, we ensure the integrity of the open-source project while leveraging IBM's
advanced DeepSearch capabilities. This strategy balances community collaboration with innovative technology,
fostering innovation and improvement in document processing for the InstructLab project.
By adopting this two-pronged approach, we ensure the integrity of the open source project while leveraging advanced DeepSearch capabilities. This strategy balances community collaboration with innovative technology,
fostering innovation and improvement in document processing for the InstructLab community.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants