Coming soon - Get a detailed view of why an account is flagged as spam!
view details

This post has been de-listed

It is no longer included in search results and normal feeds (front page, hot posts, subreddit posts, etc). It remains visible only via the author's post history.

15
AI project gaining traction -- how can I make it friendly for local models?
Post Flair (click to view more posts with a particular flair)
Post Body

Hi guys. Long story short, my project is gaining a little bit of traction on GtiHub, and I'm looking for some advice on how I can make my data extractor more friendly for local LLMs like LLaVa or QwenVL.

The library contains heuristics for extracting data from different filetypes to feed into vision-language models. Currently, I output the extracted results in the following OpenAI-friendly format:

[
  {
    "role": "user",
    "content": [
      {
        "type": "text",
        "text": "..."
      },
      {
        "type": "image_url",
        "image_url": {
          "url": "data:image/jpeg;base64,..."
        }
      }
    ]
  }

I'm assuming this won't work out-of-the-box with local models? What can I do to make my project less dependent on OpenAI?

PS: I do mention LiteLLM in the readme since I have got it working with text-only before, but I am looking for something a bit less hacky.

Thanks!

Comments
[not loaded or deleted]

This is cool, thanks for sharing!

[not loaded or deleted]

I wouldn't say it is better than langchain at the moment. Just meant to remedy the problem of langchain tools not being readily compatible with vision models :)

Author
User Disabled
Account Strength
0%
Disabled 5 months ago
Account Age
9 years
Verified Email
Yes
Verified Flair
No
Total Karma
9,205
Link Karma
7,283
Comment Karma
1,805
Profile updated: 3 days ago
Posts updated: 5 months ago

Subreddit

Post Details

We try to extract some basic information from the post title. This is not always successful or accurate, please use your best judgement and compare these values to the post title and body for confirmation.
Posted
6 months ago