github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 21:25:46 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #15171] Qwen3.5 MLX models fail to run — config.json not found in manifest #87399

New Issue
Closed
opened 2026-05-10 05:28:41 -05:00 by GiteaMirror · 10 comments
GiteaMirror commented 2026-05-10 05:28:41 -05:00
Owner
Copy Link

Originally created by @humblecoder on GitHub (Mar 31, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15171

What is the issue?

All three Qwen3.5 MLX tags from the registry pull successfully but are completely non-functional at runtime.
Affected tags:

  • qwen3.5:35b-a3b-mlx-bf16
  • qwen3.5:9b-mlx-bf16
  • qwen3.5:27b-mlx-bf16

Attempting to run (e.g. CLI, OpenWebUI, Ollama GUI) results in error:

500 Internal Server Error: mlx runner failed: Error: failed to read config.json: config "config.json" not found in manifest (exit: exit status 1)

Relevant log output

ollama pull qwen3.5:35b-a3b-mlx-bf16 && ollama pull qwen3.5:9b-mlx-bf16 && ollama pull qwen3.5:27b-mlx-bf16
pulling manifest
pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏  70 GB
writing manifest
success
⚠ Could not detect context length for qwen3.5:35b-a3b-mlx-bf16, skipping
pulling manifest
pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏  18 GB
writing manifest
success
⚠ Could not detect context length for qwen3.5:9b-mlx-bf16, skipping
pulling manifest
pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏  54 GB
writing manifest
success
⚠ Could not detect context length for qwen3.5:27b-mlx-bf16, skipping

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

ollama version is 0.19.0

Originally created by @humblecoder on GitHub (Mar 31, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15171 ### What is the issue? All three Qwen3.5 MLX tags from the registry pull successfully but are completely non-functional at runtime. Affected tags: - qwen3.5:35b-a3b-mlx-bf16 - qwen3.5:9b-mlx-bf16 - qwen3.5:27b-mlx-bf16 Attempting to run (e.g. CLI, OpenWebUI, Ollama GUI) results in error: ``` 500 Internal Server Error: mlx runner failed: Error: failed to read config.json: config "config.json" not found in manifest (exit: exit status 1) ``` ### Relevant log output ```shell ollama pull qwen3.5:35b-a3b-mlx-bf16 && ollama pull qwen3.5:9b-mlx-bf16 && ollama pull qwen3.5:27b-mlx-bf16 pulling manifest pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ 70 GB writing manifest success ⚠ Could not detect context length for qwen3.5:35b-a3b-mlx-bf16, skipping pulling manifest pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ 18 GB writing manifest success ⚠ Could not detect context length for qwen3.5:9b-mlx-bf16, skipping pulling manifest pulling model: 100% ▕██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ 54 GB writing manifest success ⚠ Could not detect context length for qwen3.5:27b-mlx-bf16, skipping ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version ollama version is 0.19.0
GiteaMirror added the bug label 2026-05-10 05:28:41 -05:00
GiteaMirror commented 2026-05-10 05:28:46 -05:00
Author
Owner
Copy Link

@CamJN commented on GitHub (Mar 31, 2026):

According to the blog post they say to use the qwen3.5:35b-a3b-bf16 model not the MLX one: https://ollama.com/blog/mlx

That doesn't make much sense given that they were announcing MLX support…

Can someone involved explain the discrepancy?

<!-- gh-comment-id:4164402770 --> @CamJN commented on GitHub (Mar 31, 2026): According to the blog post they say to use the `qwen3.5:35b-a3b-bf16` model not the MLX one: https://ollama.com/blog/mlx That doesn't make much sense given that they were announcing MLX support… Can someone involved explain the discrepancy?
GiteaMirror commented 2026-05-10 05:28:48 -05:00
Author
Owner
Copy Link

@alanina commented on GitHub (Mar 31, 2026):

I have the same issue and it's quite disappointing ...

<!-- gh-comment-id:4166119787 --> @alanina commented on GitHub (Mar 31, 2026): I have the same issue and it's quite disappointing ...
GiteaMirror commented 2026-05-10 05:28:49 -05:00
Author
Owner
Copy Link

@iahmedani commented on GitHub (Apr 1, 2026):

I'm also facing same issue

<!-- gh-comment-id:4167428109 --> @iahmedani commented on GitHub (Apr 1, 2026): I'm also facing same issue
GiteaMirror commented 2026-05-10 05:28:50 -05:00
Author
Owner
Copy Link

@saltykovdg commented on GitHub (Apr 1, 2026):

I have the same issue

<!-- gh-comment-id:4171115671 --> @saltykovdg commented on GitHub (Apr 1, 2026): I have the same issue
GiteaMirror commented 2026-05-10 05:28:51 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 1, 2026):

The blob layers in the manifests of -mlx-bf16 models are missing the name field. It's missing in all layers but since the model loader wants to access config.json for the model load, this is where the load fails.

@BruceMacD

    {                                                                                                       {
      "mediaType": "application/vnd.ollama.image.json",                                                       "mediaType": "application/vnd.ollama.image.json",
      "digest": "sha256:5e4d7f74fec2f360eb9cfbfcd6ec0c4c76e684d3a11caaed259d9fd9bfbc7944",                    "digest": "sha256:5e4d7f74fec2f360eb9cfbfcd6ec0c4c76e684d3a11caaed259d9fd9bfbc7944",
      "size": 3544,                                                                                |          "size": 3544
      "name": "config.json"                                                                        <
    },                                                                                                      },
    {                                                                                                       {
      "mediaType": "application/vnd.ollama.image.json",                                                       "mediaType": "application/vnd.ollama.image.json",
      "digest": "sha256:4f25002776b741773666203dcea8f54619f177ace3ae483d311102092a4658e0",                    "digest": "sha256:4f25002776b741773666203dcea8f54619f177ace3ae483d311102092a4658e0",
      "size": 244,                                                                                 |          "size": 244
      "name": "generation_config.json"                                                             <
    },                                                                                                      },
<!-- gh-comment-id:4173213684 --> @rick-github commented on GitHub (Apr 1, 2026): The blob layers in the manifests of -mlx-bf16 models are missing the `name` field. It's missing in all layers but since the model loader wants to access `config.json` for the model load, this is where the load fails. @BruceMacD ```json { { "mediaType": "application/vnd.ollama.image.json", "mediaType": "application/vnd.ollama.image.json", "digest": "sha256:5e4d7f74fec2f360eb9cfbfcd6ec0c4c76e684d3a11caaed259d9fd9bfbc7944", "digest": "sha256:5e4d7f74fec2f360eb9cfbfcd6ec0c4c76e684d3a11caaed259d9fd9bfbc7944", "size": 3544, | "size": 3544 "name": "config.json" < }, }, { { "mediaType": "application/vnd.ollama.image.json", "mediaType": "application/vnd.ollama.image.json", "digest": "sha256:4f25002776b741773666203dcea8f54619f177ace3ae483d311102092a4658e0", "digest": "sha256:4f25002776b741773666203dcea8f54619f177ace3ae483d311102092a4658e0", "size": 244, | "size": 244 "name": "generation_config.json" < }, }, ```
GiteaMirror commented 2026-05-10 05:28:51 -05:00
Author
Owner
Copy Link

@BruceMacD commented on GitHub (Apr 1, 2026):

Thanks for the reports, and thanks for the tag rick. This was an issue from when I re-converted the bf16 weights, I've re-uploaded them. They have the required config and should work now.

<!-- gh-comment-id:4173501546 --> @BruceMacD commented on GitHub (Apr 1, 2026): Thanks for the reports, and thanks for the tag rick. This was an issue from when I re-converted the `bf16` weights, I've re-uploaded them. They have the required config and should work now.
GiteaMirror commented 2026-05-10 05:28:52 -05:00
Author
Owner
Copy Link

@CamJN commented on GitHub (Apr 2, 2026):

@BruceMacD can you comment on why the MLX blog post didn't use the MLX models? Are they converted internally? Was the blog post wrong?

<!-- gh-comment-id:4180216618 --> @CamJN commented on GitHub (Apr 2, 2026): @BruceMacD can you comment on why the MLX blog post didn't use the MLX models? Are they converted internally? Was the blog post wrong?
GiteaMirror commented 2026-05-10 05:28:52 -05:00
Author
Owner
Copy Link

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

The MLX models are the ones with a quantization tag of int4, int8, mxfp4, nvfp4 or mlx-bf16. The model details page of a MLX model will show a lot of JSON structures rather than the single model line of a GGUF model.

<!-- gh-comment-id:4181028576 --> @rick-github commented on GitHub (Apr 2, 2026): The MLX models are the ones with a quantization tag of `int4`, `int8`, `mxfp4`, `nvfp4` or `mlx-bf16`. The [model details](https://ollama.com/library/qwen3.5:2b-mxfp8) page of a MLX model will show a lot of JSON structures rather than the single `model` line of a [GGUF model](https://ollama.com/library/qwen3.5:0.8b-bf16).
GiteaMirror commented 2026-05-10 05:28:53 -05:00
Author
Owner
Copy Link

@CamJN commented on GitHub (Apr 2, 2026):

@rick-github so qwen3.5:35b-a3b-bf16 used in the blog post is not an mlx model given those criteria, so the blog post should be corrected, no?

<!-- gh-comment-id:4181056237 --> @CamJN commented on GitHub (Apr 2, 2026): @rick-github so `qwen3.5:35b-a3b-bf16` used in the blog post is not an mlx model given those criteria, so the blog post should be corrected, no?
GiteaMirror commented 2026-05-10 05:28:54 -05:00
Author
Owner
Copy Link

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

The blog post doesn't mention qwen3.5:35b-a3b-bf16, it uses qwen3.5:35b-a3b-coding-nvfp4 in the examples.

<!-- gh-comment-id:4181074005 --> @rick-github commented on GitHub (Apr 3, 2026): The blog post doesn't mention `qwen3.5:35b-a3b-bf16`, it uses `qwen3.5:35b-a3b-coding-nvfp4` in the examples.
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#87399
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?