github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #14053] Ollama glm-ocr returning blank Markdown on Mac #9183

New Issue
Closed
opened 2026-04-12 22:01:58 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-12 22:01:58 -05:00
Owner
Copy Link

Originally created by @githycody on GitHub (Feb 3, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14053

What is the issue?

When I run the the model with the image:
ollama run glm-ocr Text Recognition:

it returns this:


```markdown

```markdown

```text

I have been in contact with people from the team, who are able to make it work. Unfortunately only on Windows. Perhaps the issue can be reproduced on Mac.

ollama ps
NAME ID SIZE PROCESSOR CONTEXT UNTIL
glm-ocr:latest 6effedd0dc8a 3.8 GB 100% GPU 4096 4 minutes from now

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

ollama version is 0.15.5-rc1

Originally created by @githycody on GitHub (Feb 3, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14053 ### What is the issue? When I run the the model with the image: ollama run glm-ocr Text Recognition: <imagePath> it returns this: ```markdown ```markdown ```markdown ```text ``` I have been in contact with people from the team, who are able to make it work. Unfortunately only on Windows. Perhaps the issue can be reproduced on Mac. ollama ps NAME ID SIZE PROCESSOR CONTEXT UNTIL glm-ocr:latest 6effedd0dc8a 3.8 GB 100% GPU 4096 4 minutes from now ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version ollama version is 0.15.5-rc1
GiteaMirror added the bug label 2026-04-12 22:01:58 -05:00
GiteaMirror commented 2026-04-12 22:02:00 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 3, 2026):

For reference: https://discord.com/channels/1128867683291627614/1468279523496956025

It works on Linux on both Nvidia and ROCm devices.

<!-- gh-comment-id:3843674439 --> @rick-github commented on GitHub (Feb 3, 2026): For reference: https://discord.com/channels/1128867683291627614/1468279523496956025 It works on Linux on both Nvidia and ROCm devices.
GiteaMirror commented 2026-04-12 22:02:00 -05:00
Author
Owner
Copy Link

@starchaser01 commented on GitHub (Feb 3, 2026):

I had the same problem. After reducing the image resolution, it worked.

<!-- gh-comment-id:3844265651 --> @starchaser01 commented on GitHub (Feb 3, 2026): I had the same problem. After reducing the image resolution, it worked.
GiteaMirror commented 2026-04-12 22:02:01 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 4, 2026):

Agreed, this looks like a resolution issue. An A4 page at about 150dpi is the largest image the model can handle.

<!-- gh-comment-id:3844783818 --> @rick-github commented on GitHub (Feb 4, 2026): Agreed, this looks like a resolution issue. An A4 page at about 150dpi is the largest image the model can handle.
GiteaMirror commented 2026-04-12 22:02:02 -05:00
Author
Owner
Copy Link

@githycody commented on GitHub (Feb 4, 2026):

@rick-github but do you think this an Ollama problem? the specs specify 71mp.

<!-- gh-comment-id:3845913233 --> @githycody commented on GitHub (Feb 4, 2026): @rick-github but do you think this an Ollama problem? the specs specify 71mp.
GiteaMirror commented 2026-04-12 22:02:02 -05:00
Author
Owner
Copy Link

@griff12 commented on GitHub (Feb 5, 2026):

I am having the same problem on Windows as well.

<!-- gh-comment-id:3850547403 --> @griff12 commented on GitHub (Feb 5, 2026): I am having the same problem on Windows as well.
GiteaMirror commented 2026-04-12 22:02:03 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 5, 2026):

The specs may specify 71mp, but looking at the SDK it looks like page OCR is 1mp by default and the image is scaled to 200 dpi. There's also a smart resizer which by default limits images to 12mp. I'm not familiar with the SDK though, maybe there's some sort of image partitioner that stitches the results together.

In any case, the ollama glmocr image processor has a default if 3.2mp, which is about the size I experimentally reached above. The comment indicates that this is to keep memory stable, whatever that means. It's theoretically possible to change this by adjusting KV values in the model, but that requires editing the GGUF file, which is a pain. I'll experiment and see what I can come up with. In the meantime, resizing images to less than 3mp is the best way to allow glm-ocr to return useful results.

<!-- gh-comment-id:3850727098 --> @rick-github commented on GitHub (Feb 5, 2026): The specs may specify 71mp, but looking at the [SDK](https://github.com/zai-org/GLM-OCR) it looks like page OCR is [1mp](https://github.com/zai-org/GLM-OCR/blob/8e6fd70a31cfbd072746a9a2f45c178efc644048/glmocr/config.py#L110) by default and the image is scaled to [200 dpi](https://github.com/zai-org/GLM-OCR/blob/8e6fd70a31cfbd072746a9a2f45c178efc644048/glmocr/config.py#L119). There's also a [smart resizer](https://github.com/zai-org/GLM-OCR/blob/8e6fd70a31cfbd072746a9a2f45c178efc644048/glmocr/utils/image_utils.py#L13) which by default limits images to 12mp. I'm not familiar with the SDK though, maybe there's some sort of image partitioner that stitches the results together. In any case, the ollama glmocr [image processor](https://github.com/ollama/ollama/blob/c61023f5548f61651b7fd04393e2a93430f89a71/model/models/glmocr/imageprocessor.go#L33) has a default if 3.2mp, which is about the size I experimentally reached above. The comment indicates that this is to keep memory stable, whatever that means. It's theoretically possible to change this by adjusting KV values in the model, but that requires editing the GGUF file, which is a pain. I'll experiment and see what I can come up with. In the meantime, resizing images to less than 3mp is the best way to allow glm-ocr to return useful results.
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 bug
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#9183
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?