github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 00:51:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13896] Support for Macbook m5 for image generation (Unable to load kernel affine_qmm…bfloat16…) #55605

New Issue
Closed
opened 2026-04-29 09:28:30 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-29 09:28:30 -05:00
Owner
Copy Link

Originally created by @anthropoleo on GitHub (Jan 25, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/13896

What is the issue?

Image generation fails on Apple M5 when using the Ollama image runner with model x/flux2-klein:4b. The image runner starts successfully, loads the model, then fails during “Evaluating setup” with an MLX Metal kernel load error:

MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_4_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0

The request returns HTTP 500 from POST /api/generate.

Environment
• Hardware: Apple M5 (Metal)
• GPU detected: Metal description="Apple M5" total VRAM 17.8 GiB
• Ollama version: 0.15.0
• Host: macOS app bundle (/Applications/Ollama.app/...)
• Model: x/flux2-klein:4b
• Image output size: 1024x1024
• Runner backend: MLX (Metal)

Steps to reproduce
1. Start Ollama (macOS app)
2. Confirm server is listening on 127.0.0.1:11434
3. Trigger image generation with x/flux2-klein:4b (for example via POST /api/generate)
4. Observe request fails with 500, and the image-runner logs show the MLX Metal kernel load error.

Generation fails quickly during setup. Server returns HTTP 500. Image runner logs show MLX Metal kernel load error for a bfloat16 affine_qmm kernel.

Relevant log output

time=2026-01-24T20:10:06.840+10:00 level=INFO source=routes.go:1631 msg="server config" env="map[... OLLAMA_HOST:http://127.0.0.1:11434 ...]"
time=2026-01-24T20:10:06.880+10:00 level=INFO source=routes.go:1684 msg="Listening on 127.0.0.1:11434 (version 0.15.0)"
time=2026-01-24T20:10:06.880+10:00 level=INFO source=runner.go:67 msg="discovering available GPUs..."
time=2026-01-24T20:10:12.803+10:00 level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=Metal compute=0.0 name=Metal description="Apple M5" libdirs="" driver=0.0 pci_id="" type=discrete total="17.8 GiB" available="17.8 GiB"
time=2026-01-24T20:10:12.803+10:00 level=INFO source=routes.go:1725 msg="entering low vram mode" "total vram"="17.8 GiB" threshold="20.0 GiB"

[GIN] 2026/01/25 - 13:02:06 | 200 |   50.156375ms | 127.0.0.1 | POST "/api/show"
time=2026-01-25T13:02:06.798+10:00 level=INFO source=server.go:143 msg="starting image runner subprocess" exe=/Applications/Ollama.app/Contents/Resources/ollama model=x/flux2-klein:4b port=55359
time=2026-01-25T13:02:06.831+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:06 runner.go:87: INFO MLX library initialized"
time=2026-01-25T13:02:06.833+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:06 runner.go:92: INFO detected model type type=Flux2KleinPipeline"
time=2026-01-25T13:02:06.833+10:00 level=INFO source=server.go:129 msg=image-runner msg="Loading FLUX.2 Klein model from manifest: x/flux2-klein:4b..."
time=2026-01-25T13:02:06.973+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Loading tokenizer... ✓"
time=2026-01-25T13:02:07.798+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Loading text encoder... ✓"
time=2026-01-25T13:02:08.508+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Loading transformer... ✓"
time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Loading VAE... ✓"
time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Evaluating weights... ✓"
time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Loaded in 1.75s (5.3 GB VRAM)"
time=2026-01-25T13:02:08.588+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:08 runner.go:139: INFO image runner listening addr=127.0.0.1:55359"
time=2026-01-25T13:02:08.600+10:00 level=INFO source=server.go:214 msg="image runner is ready" port=55359
[GIN] 2026/01/25 - 13:02:08 | 200 | 1.843952958s | 127.0.0.1 | POST "/api/generate"

time=2026-01-25T13:02:20.600+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Output: 1024x1024"
time=2026-01-25T13:02:20.603+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Encoding prompt... ✓"
time=2026-01-25T13:02:20.674+10:00 level=INFO source=server.go:129 msg=image-runner msg="  Evaluating setup... MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_4_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0"
time=2026-01-25T13:02:20.674+10:00 level=INFO source=server.go:129 msg=image-runner msg=" at /Users/runner/work/ollama/ollama/build/_deps/mlx-c-src/mlx/c/transforms.cpp:73"
[GIN] 2026/01/25 - 13:02:20 | 500 | 263.07925ms | 127.0.0.1 | POST "/api/generate"

OS

No response

GPU

No response

CPU

No response

Ollama version

No response

Originally created by @anthropoleo on GitHub (Jan 25, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/13896 ### What is the issue? Image generation fails on Apple M5 when using the Ollama image runner with model x/flux2-klein:4b. The image runner starts successfully, loads the model, then fails during “Evaluating setup” with an MLX Metal kernel load error: MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_4_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0 The request returns HTTP 500 from POST /api/generate. Environment • Hardware: Apple M5 (Metal) • GPU detected: Metal description="Apple M5" total VRAM 17.8 GiB • Ollama version: 0.15.0 • Host: macOS app bundle (/Applications/Ollama.app/...) • Model: x/flux2-klein:4b • Image output size: 1024x1024 • Runner backend: MLX (Metal) Steps to reproduce 1. Start Ollama (macOS app) 2. Confirm server is listening on 127.0.0.1:11434 3. Trigger image generation with x/flux2-klein:4b (for example via POST /api/generate) 4. Observe request fails with 500, and the image-runner logs show the MLX Metal kernel load error. Generation fails quickly during setup. Server returns HTTP 500. Image runner logs show MLX Metal kernel load error for a bfloat16 affine_qmm kernel. ### Relevant log output ```shell time=2026-01-24T20:10:06.840+10:00 level=INFO source=routes.go:1631 msg="server config" env="map[... OLLAMA_HOST:http://127.0.0.1:11434 ...]" time=2026-01-24T20:10:06.880+10:00 level=INFO source=routes.go:1684 msg="Listening on 127.0.0.1:11434 (version 0.15.0)" time=2026-01-24T20:10:06.880+10:00 level=INFO source=runner.go:67 msg="discovering available GPUs..." time=2026-01-24T20:10:12.803+10:00 level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=Metal compute=0.0 name=Metal description="Apple M5" libdirs="" driver=0.0 pci_id="" type=discrete total="17.8 GiB" available="17.8 GiB" time=2026-01-24T20:10:12.803+10:00 level=INFO source=routes.go:1725 msg="entering low vram mode" "total vram"="17.8 GiB" threshold="20.0 GiB" [GIN] 2026/01/25 - 13:02:06 | 200 | 50.156375ms | 127.0.0.1 | POST "/api/show" time=2026-01-25T13:02:06.798+10:00 level=INFO source=server.go:143 msg="starting image runner subprocess" exe=/Applications/Ollama.app/Contents/Resources/ollama model=x/flux2-klein:4b port=55359 time=2026-01-25T13:02:06.831+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:06 runner.go:87: INFO MLX library initialized" time=2026-01-25T13:02:06.833+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:06 runner.go:92: INFO detected model type type=Flux2KleinPipeline" time=2026-01-25T13:02:06.833+10:00 level=INFO source=server.go:129 msg=image-runner msg="Loading FLUX.2 Klein model from manifest: x/flux2-klein:4b..." time=2026-01-25T13:02:06.973+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Loading tokenizer... ✓" time=2026-01-25T13:02:07.798+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Loading text encoder... ✓" time=2026-01-25T13:02:08.508+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Loading transformer... ✓" time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Loading VAE... ✓" time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Evaluating weights... ✓" time=2026-01-25T13:02:08.587+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Loaded in 1.75s (5.3 GB VRAM)" time=2026-01-25T13:02:08.588+10:00 level=WARN source=server.go:136 msg=image-runner msg="2026/01/25 13:02:08 runner.go:139: INFO image runner listening addr=127.0.0.1:55359" time=2026-01-25T13:02:08.600+10:00 level=INFO source=server.go:214 msg="image runner is ready" port=55359 [GIN] 2026/01/25 - 13:02:08 | 200 | 1.843952958s | 127.0.0.1 | POST "/api/generate" time=2026-01-25T13:02:20.600+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Output: 1024x1024" time=2026-01-25T13:02:20.603+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Encoding prompt... ✓" time=2026-01-25T13:02:20.674+10:00 level=INFO source=server.go:129 msg=image-runner msg=" Evaluating setup... MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_4_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0" time=2026-01-25T13:02:20.674+10:00 level=INFO source=server.go:129 msg=image-runner msg=" at /Users/runner/work/ollama/ollama/build/_deps/mlx-c-src/mlx/c/transforms.cpp:73" [GIN] 2026/01/25 - 13:02:20 | 500 | 263.07925ms | 127.0.0.1 | POST "/api/generate" ``` ### OS _No response_ ### GPU _No response_ ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the bug label 2026-04-29 09:28:30 -05:00
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#55605
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?