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 #11217] image size of qwen2.5-vl #7390

New Issue
Closed
opened 2026-04-12 19:28:41 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-12 19:28:41 -05:00
Owner
Copy Link

Originally created by @wubangcai on GitHub (Jun 27, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11217

When I deployed qwen2.5-vl-72B using ollama and input the image for inference, to what scale will the image be scaled at this time? When I directly asked the model about the image size, the responses showed 10001000 and 1200800. However, I know that the image size of qwen2.5-vl must be an integer multiple of 28. Why is that?

Originally created by @wubangcai on GitHub (Jun 27, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11217 When I deployed qwen2.5-vl-72B using ollama and input the image for inference, to what scale will the image be scaled at this time? When I directly asked the model about the image size, the responses showed 1000*1000 and 1200*800. However, I know that the image size of qwen2.5-vl must be an integer multiple of 28. Why is that?
GiteaMirror added the question label 2026-04-12 19:28:41 -05:00
GiteaMirror commented 2026-04-12 19:28:42 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 27, 2025):

qwne2.5vl resizes images such that the dimensions are a multiple of 28, and the final image has less than 1M pixels. So 1200x800 is resized to 1204x812, and 1000x1000 is resized to 980x980. Unless models are explicitly trained with the information, they generally can't answer questions about model parameters, as they have no introspection. The value of 28 is derived from the model parameters of patch_size and spatial_merge_size which control how the vision processor handles the images. For qwen2.5vl, patch_size is 14, spatial_merge_size is 2, and this constrains the processing to patch_size * spatial_merge_size or 28 pixels.

<!-- gh-comment-id:3013302452 --> @rick-github commented on GitHub (Jun 27, 2025): qwne2.5vl resizes images such that the dimensions are a multiple of 28, and the final image has less than 1M pixels. So 1200x800 is resized to 1204x812, and 1000x1000 is resized to 980x980. Unless models are explicitly trained with the information, they generally can't answer questions about model parameters, as they have no introspection. The value of 28 is derived from the model parameters of `patch_size` and `spatial_merge_size` which control how the vision processor handles the images. For qwen2.5vl, `patch_size` is 14, `spatial_merge_size` is 2, and this constrains the processing to `patch_size` * `spatial_merge_size` or 28 pixels.
GiteaMirror commented 2026-04-12 19:28:42 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Jul 1, 2025):

Going to mark this as answered. (Thank you @rick-github !)

<!-- gh-comment-id:3025058486 --> @pdevine commented on GitHub (Jul 1, 2025): Going to mark this as answered. (Thank you @rick-github !)
GiteaMirror commented 2026-04-12 19:28:42 -05:00
Author
Owner
Copy Link

@wubangcai commented on GitHub (Jul 8, 2025):

If I want the size of the input image to exceed 1M, how should I set it? @rick-github

<!-- gh-comment-id:3047563500 --> @wubangcai commented on GitHub (Jul 8, 2025): If I want the size of the input image to exceed 1M, how should I set it? @rick-github
GiteaMirror commented 2026-04-12 19:28:42 -05:00
Author
Owner
Copy Link

@crackerfly commented on GitHub (Jul 29, 2025):

qwne2.5vl resizes images such that the dimensions are a multiple of 28, and the final image has less than 1M pixels. So 1200x800 is resized to 1204x812, and 1000x1000 is resized to 980x980. Unless models are explicitly trained with the information, they generally can't answer questions about model parameters, as they have no introspection. The value of 28 is derived from the model parameters of patch_size and spatial_merge_size which control how the vision processor handles the images. For qwen2.5vl, patch_size is 14, spatial_merge_size is 2, and this constrains the processing to patch_size * spatial_merge_size or 28 pixels.

Hi～
ollama: 0.9.3
model: qwen2.5vl:7b-q8_0
question: i entered a image resolution of 798x519, and ollama reported an error directly, but when I crop the image resolution to 798x518, it works fine, what is the reason for this

<!-- gh-comment-id:3132036966 --> @crackerfly commented on GitHub (Jul 29, 2025): > qwne2.5vl resizes images such that the dimensions are a multiple of 28, and the final image has less than 1M pixels. So 1200x800 is resized to 1204x812, and 1000x1000 is resized to 980x980. Unless models are explicitly trained with the information, they generally can't answer questions about model parameters, as they have no introspection. The value of 28 is derived from the model parameters of `patch_size` and `spatial_merge_size` which control how the vision processor handles the images. For qwen2.5vl, `patch_size` is 14, `spatial_merge_size` is 2, and this constrains the processing to `patch_size` * `spatial_merge_size` or 28 pixels. Hi～ ollama: 0.9.3 model: qwen2.5vl:7b-q8_0 question: i entered a image resolution of 798x519, and ollama reported an error directly, but when I crop the image resolution to 798x518, it works fine, what is the reason for this
GiteaMirror commented 2026-04-12 19:28:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 29, 2025):

Error?

<!-- gh-comment-id:3132122180 --> @rick-github commented on GitHub (Jul 29, 2025): Error?
GiteaMirror commented 2026-04-12 19:28:43 -05:00
Author
Owner
Copy Link

@crackerfly commented on GitHub (Jul 30, 2025):

Error?

I can't reproduce the problem after restarting the operating system...

<!-- gh-comment-id:3134552628 --> @crackerfly commented on GitHub (Jul 30, 2025): > Error? I can't reproduce the problem after restarting the operating system...
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 question
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#7390
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?