[GH-ISSUE #9780] How to use a multimodal model back in ollama? #6394

Open
opened 2026-04-12 17:54:30 -05:00 by GiteaMirror · 3 comments
Owner

Originally created by @KansaiTraining on GitHub (Mar 15, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9780

I use ollama to run several models, including multimodal ones such as Llava.

In my organization, we have fine tuned llava for our particular task, and as a result we got two gguf fies (one for the vision part and one for the text part).

I know how to create a model from one gguf file as detailed in the documentation but how can I create a model to run with ollama from the result of fine tuned multimodals?

Originally created by @KansaiTraining on GitHub (Mar 15, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9780 I use ollama to run several models, including multimodal ones such as Llava. In my organization, we have fine tuned llava for our particular task, and as a result we got two gguf fies (one for the vision part and one for the text part). I know how to create a model from one gguf file as detailed in the [documentation ](https://github.com/ollama/ollama/blob/main/docs/modelfile.md) but how can I create a model to run with ollama from the result of fine tuned multimodals?
Author
Owner

@EntropyYue commented on GitHub (Mar 16, 2025):

Import the model and projector separately like this

FROM model.gguf
FROM projector.gguf
<!-- gh-comment-id:2727167820 --> @EntropyYue commented on GitHub (Mar 16, 2025): Import the model and projector separately like this ```modelfile FROM model.gguf FROM projector.gguf ```
Author
Owner

@KansaiTraining commented on GitHub (Mar 16, 2025):

Import the model and projector separately like this

FROM model.gguf
FROM projector.gguf

will this work?
I suppose I will give it a try ...

<!-- gh-comment-id:2727364131 --> @KansaiTraining commented on GitHub (Mar 16, 2025): > Import the model and projector separately like this > > ``` > FROM model.gguf > FROM projector.gguf > ``` will this work? I suppose I will give it a try ...
Author
Owner

@rjmalagon commented on GitHub (Mar 16, 2025):

Import the model and projector separately like this

FROM model.gguf
FROM projector.gguf

This is correct @KansaiTraining, I was struggling with this because the documentation is scarce and there is wrong info about this on the GitHub issues too (easy for newcomers to get confused with the adapters part of the model file when trying to add the projector).

The proper "fix" to this issue is update documentation on how to properly add the additional components of multimodal models.

<!-- gh-comment-id:2727549778 --> @rjmalagon commented on GitHub (Mar 16, 2025): > Import the model and projector separately like this > > ``` > FROM model.gguf > FROM projector.gguf > ``` This is correct @KansaiTraining, I was struggling with this because the documentation is scarce and there is wrong info about this on the GitHub issues too (easy for newcomers to get confused with the adapters part of the model file when trying to add the projector). The proper "fix" to this issue is update documentation on how to properly add the additional components of multimodal models.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#6394