github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13594] grounding coordinates #34711

New Issue
Open
opened 2026-04-22 18:29:06 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 18:29:06 -05:00
Owner
Copy Link

Originally created by @Abdulrahman392011 on GitHub (Jan 1, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/13594

is there a way to use grounding with vision models other than just prompt engineering. i tried using structured output but it didn't work properly with moondream:1.8b.

i know that some models are good with detecting triggering words in the prompt and bring out the coordinates but it's not really reliable method and it's kinda hit or miss.

i'm interested in the smaller models, the big ones have no issues detecting the trigger words. it's usually the smaller ones that give me a hard time.

Originally created by @Abdulrahman392011 on GitHub (Jan 1, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/13594 is there a way to use grounding with vision models other than just prompt engineering. i tried using structured output but it didn't work properly with moondream:1.8b. i know that some models are good with detecting triggering words in the prompt and bring out the coordinates but it's not really reliable method and it's kinda hit or miss. i'm interested in the smaller models, the big ones have no issues detecting the trigger words. it's usually the smaller ones that give me a hard time.
GiteaMirror added the feature request label 2026-04-22 18:29:06 -05:00
GiteaMirror commented 2026-04-22 18:29:07 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 2, 2026):

This is model dependent. For moondream, the model author suggests prompting with Bounding box: {object}

<!-- gh-comment-id:3704256780 --> @rick-github commented on GitHub (Jan 2, 2026): This is model dependent. For moondream, the model author suggests prompting with [`Bounding box: {object}`](https://github.com/vikhyat/moondream/issues/101)
GiteaMirror commented 2026-04-22 18:29:08 -05:00
Author
Owner
Copy Link

@Abdulrahman392011 commented on GitHub (Jan 2, 2026):

wouldn't it be a good idea to have ollama api deal with the discrepancies and needs of each model. and users have a somewhat unified way of using the models through the api.

on a separate note why is there no small vision models in ollama. something like florence-2 for instance is 0.2b for the base and 0.8b for the large size and it works well on edge devices. moondream is 1.8b and that's the smallest one i can find on ollama.

<!-- gh-comment-id:3705075263 --> @Abdulrahman392011 commented on GitHub (Jan 2, 2026): wouldn't it be a good idea to have ollama api deal with the discrepancies and needs of each model. and users have a somewhat unified way of using the models through the api. on a separate note why is there no small vision models in ollama. something like florence-2 for instance is 0.2b for the base and 0.8b for the large size and it works well on edge devices. moondream is 1.8b and that's the smallest one i can find on ollama.
GiteaMirror commented 2026-04-22 18:29:09 -05:00
Author
Owner
Copy Link

@SingularityMan commented on GitHub (Jan 20, 2026):

wouldn't it be a good idea to have ollama api deal with the discrepancies and needs of each model. and users have a somewhat unified way of using the models through the api.

on a separate note why is there no small vision models in ollama. something like florence-2 for instance is 0.2b for the base and 0.8b for the large size and it works well on edge devices. moondream is 1.8b and that's the smallest one i can find on ollama.

You can try qwen3vl-2b but you need to scale each image to 1000x1000 because that's the image resolution it was trained for. Every model is different and Ollama won't be able to accommodate all of them.

<!-- gh-comment-id:3773944779 --> @SingularityMan commented on GitHub (Jan 20, 2026): > wouldn't it be a good idea to have ollama api deal with the discrepancies and needs of each model. and users have a somewhat unified way of using the models through the api. > > on a separate note why is there no small vision models in ollama. something like florence-2 for instance is 0.2b for the base and 0.8b for the large size and it works well on edge devices. moondream is 1.8b and that's the smallest one i can find on ollama. You can try qwen3vl-2b but you need to scale each image to 1000x1000 because that's the image resolution it was trained for. Every model is different and Ollama won't be able to accommodate all of them.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label feature request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#34711
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?