github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 05:05:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #16021] bug: x/flux2-klein silently returns empty response in <10ms on Apple Silicon, DMG v0.23.1 #87889

New Issue
Open
opened 2026-05-10 06:32:12 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-05-10 06:32:12 -05:00
Owner
Copy Link

Originally created by @AMatisse on GitHub (May 6, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/16021

What is the issue?

Describe the bug

On Ollama v0.23.1 (macOS DMG install), running x/flux2-klein on Apple Silicon
completes instantly with no error and no image produced. The API returns 4 progress
steps in under 10ms with empty response fields, then stops. No file is written to
disk, no error message is shown in the CLI or logs.

Environment

  • Ollama version: 0.23.1
  • Install method: DMG (official, from ollama.com)
  • OS: macOS (Apple Silicon, arm64)
  • Model: x/flux2-klein (FP4, 8.0B)

Steps to reproduce

  1. Run ollama run x/flux2-klein "a cat in the rain"
  2. The command exits immediately with no output and no image.

Or via API:

curl -s http://localhost:11434/api/generate \
  -d '{"model":"x/flux2-klein","prompt":"a cat in the rain"}'

Actual API response

{"model":"x/flux2-klein","response":"","done":false,"total":4}
{"model":"x/flux2-klein","response":"","done":false,"completed":1,"total":4}
{"model":"x/flux2-klein","response":"","done":false,"completed":2,"total":4}
{"model":"x/flux2-klein","response":"","done":false,"completed":3,"total":4}
{"model":"x/flux2-klein","response":"","done":false,"completed":4,"total":4}

All 4 steps complete in under 10ms. The response field is always empty.
No image is saved anywhere on disk.

Expected behavior

Image generation should take 30–60 seconds and produce a .png file in the
current directory, as documented.

Bundle investigation

No dedicated image runner binary exists in the DMG bundle:

find /Applications/Ollama.app -name "*runner*" -o -name "*image*"
# → no output

The only runner is the main ollama binary. The mlx_metal_v3 and mlx_metal_v4
directories contain MLX dylibs, but all other bundled GPU runners are x86 CPU
variants (libggml-cpu-haswell.so, sandybridge, etc.) — none for Apple Silicon
image generation. No MLX runner subprocess is started when generation is triggered.

Additional context

  • No log file at ~/Library/Logs/Ollama/server.log
  • This is distinct from #15882 (Homebrew-specific) and #15456 (x86_64 dylib issue)
  • This occurs on a clean DMG install with zero error output

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.23.1

Originally created by @AMatisse on GitHub (May 6, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/16021 ### What is the issue? ## Describe the bug On Ollama v0.23.1 (macOS DMG install), running `x/flux2-klein` on Apple Silicon completes instantly with no error and no image produced. The API returns 4 progress steps in under 10ms with empty `response` fields, then stops. No file is written to disk, no error message is shown in the CLI or logs. ## Environment - **Ollama version:** 0.23.1 - **Install method:** DMG (official, from ollama.com) - **OS:** macOS (Apple Silicon, arm64) - **Model:** x/flux2-klein (FP4, 8.0B) ## Steps to reproduce 1. Run `ollama run x/flux2-klein "a cat in the rain"` 2. The command exits immediately with no output and no image. Or via API: curl -s http://localhost:11434/api/generate \ -d '{"model":"x/flux2-klein","prompt":"a cat in the rain"}' ## Actual API response {"model":"x/flux2-klein","response":"","done":false,"total":4} {"model":"x/flux2-klein","response":"","done":false,"completed":1,"total":4} {"model":"x/flux2-klein","response":"","done":false,"completed":2,"total":4} {"model":"x/flux2-klein","response":"","done":false,"completed":3,"total":4} {"model":"x/flux2-klein","response":"","done":false,"completed":4,"total":4} All 4 steps complete in under 10ms. The `response` field is always empty. No image is saved anywhere on disk. ## Expected behavior Image generation should take 30–60 seconds and produce a `.png` file in the current directory, as documented. ## Bundle investigation No dedicated image runner binary exists in the DMG bundle: find /Applications/Ollama.app -name "*runner*" -o -name "*image*" # → no output The only runner is the main `ollama` binary. The `mlx_metal_v3` and `mlx_metal_v4` directories contain MLX dylibs, but all other bundled GPU runners are x86 CPU variants (`libggml-cpu-haswell.so`, `sandybridge`, etc.) — none for Apple Silicon image generation. No MLX runner subprocess is started when generation is triggered. ## Additional context - No log file at `~/Library/Logs/Ollama/server.log` - This is distinct from #15882 (Homebrew-specific) and #15456 (x86_64 dylib issue) - This occurs on a clean DMG install with zero error output ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.23.1
GiteaMirror added the bug label 2026-05-10 06:32:12 -05:00
GiteaMirror commented 2026-05-10 06:32:46 -05:00
Author
Owner
Copy Link

@EVSalomon commented on GitHub (May 6, 2026):

I can confirm it. The same is happening with z-image-turbo.

<!-- gh-comment-id:4391735417 --> @EVSalomon commented on GitHub (May 6, 2026): I can confirm it. The same is happening with z-image-turbo.
GiteaMirror commented 2026-05-10 06:32:49 -05:00
Author
Owner
Copy Link

@ed-norris commented on GitHub (May 6, 2026):

Both of these models worked as expected with 0.23.0

<!-- gh-comment-id:4392632466 --> @ed-norris commented on GitHub (May 6, 2026): Both of these models worked as expected with 0.23.0
GiteaMirror commented 2026-05-10 06:32:51 -05:00
Author
Owner
Copy Link

@CHINMAYVIVEK commented on GitHub (May 9, 2026):

I can confirm this. Ollama versions after 0.23.0 are not working with x/z-image-turbo and x/flux2-klein.
Both models worked as expected on version 0.23.0, but in later versions the process completes immediately without any error logs or response.

<!-- gh-comment-id:4410719183 --> @CHINMAYVIVEK commented on GitHub (May 9, 2026): I can confirm this. Ollama versions after 0.23.0 are not working with `x/z-image-turbo` and `x/flux2-klein`. Both models worked as expected on version 0.23.0, but in later versions the process completes immediately without any error logs or response.
GiteaMirror commented 2026-05-10 06:32:53 -05:00
Author
Owner
Copy Link

@sergio-ildefonso commented on GitHub (May 9, 2026):

This was what I did ( on my mac ), and it worked for me:

Uninstall Ollama and install the previous working version (0.23.0)

Uninstall

Close the application

killall Ollama

Remove the application and executable

sudo rm -rf /Applications/Ollama.app
sudo rm /usr/local/bin/ollama

Clear the cache and support data

rm -rf ~/Library/Application\ Support/Ollama
rm -rf ~/Library/Caches/com.electron.ollama
rm -rf ~/Library/Saved\ Application\ State/com.electron.ollama.savedState

(Optional) Delete downloaded templates

rm -rf ~/.ollama

Install

1. Download the official Ollama binary file from GitHub

curl -L -o Ollama-darwin.zip github.com

(Alternatively, download from the repository: https://github.com/ollama/ollama/releases/download/v0.23.0/Ollama-darwin.zip)

2. Unzip directly into the Applications folder on your Mac

unzip Ollama-darwin.zip -d /Applications/

(Alternatively, unzip the file and place the executable in the Applications folder)

3. Delete the temporary zip file to save disk space

rm Ollama-darwin.zip

4. Create the symbolic link to map the 'ollama' command Globally in Terminal

sudo ln -sf /Applications/Ollama.app/Contents/Resources/ollama /usr/local/bin/ollama

5. Remove macOS quarantine to avoid "Untrusted App" blocking

xattr -r -d com.apple.quarantine /Applications/Ollama.app
<!-- gh-comment-id:4413382398 --> @sergio-ildefonso commented on GitHub (May 9, 2026): This was what I did ( on my mac ), and it worked for me: # Uninstall Ollama and install the previous working version (0.23.0) ## Uninstall ### Close the application ```bash killall Ollama ``` ### Remove the application and executable ```bash sudo rm -rf /Applications/Ollama.app sudo rm /usr/local/bin/ollama ``` ### Clear the cache and support data ```bash rm -rf ~/Library/Application\ Support/Ollama rm -rf ~/Library/Caches/com.electron.ollama rm -rf ~/Library/Saved\ Application\ State/com.electron.ollama.savedState ``` ### (Optional) Delete downloaded templates ```bash rm -rf ~/.ollama ``` ## Install ## 1. Download the official Ollama binary file from GitHub ```bash curl -L -o Ollama-darwin.zip github.com ``` (Alternatively, download from the repository: https://github.com/ollama/ollama/releases/download/v0.23.0/Ollama-darwin.zip) ## 2. Unzip directly into the Applications folder on your Mac ```bash unzip Ollama-darwin.zip -d /Applications/ ``` (Alternatively, unzip the file and place the executable in the Applications folder) ## 3. Delete the temporary zip file to save disk space ```bash rm Ollama-darwin.zip ``` ## 4. Create the symbolic link to map the 'ollama' command Globally in Terminal ```bash sudo ln -sf /Applications/Ollama.app/Contents/Resources/ollama /usr/local/bin/ollama ``` ## 5. Remove macOS quarantine to avoid "Untrusted App" blocking ```bash xattr -r -d com.apple.quarantine /Applications/Ollama.app ```
GiteaMirror commented 2026-05-10 06:32:55 -05:00
Author
Owner
Copy Link

@ed-norris commented on GitHub (May 9, 2026):

That seems like overkill - just drag the Ollama app to the trash and get the new .dmg from https://github.com/ollama/ollama/releases/tag/v0.23.0

<!-- gh-comment-id:4413737591 --> @ed-norris commented on GitHub (May 9, 2026): That seems like overkill - just drag the Ollama app to the trash and get the new .dmg from https://github.com/ollama/ollama/releases/tag/v0.23.0
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#87889
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?