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 #14118] MLX Error #34971

New Issue
Open
opened 2026-04-22 19:03:56 -05:00 by GiteaMirror · 24 comments
GiteaMirror commented 2026-04-22 19:03:56 -05:00
Owner
Copy Link

Originally created by @hwitzthum on GitHub (Feb 6, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14118

What is the issue?

Config:
macOS version (26.2)
chip (M5)
Ollama version (0.15.5)
The MLX kernel error: MLX error: [metal::Device] Unable to load kernel.

The model loads successfully into VRAM (11.9 GB), but when it tries to actually generate the image, it fails to load a Metal GPU kernel

Relevant log output

time=2026-02-06T11:22:05.172+01:00 level=INFO source=server.go:219 msg="mlx runner is ready" port=50052
time=2026-02-06T11:22:05.270+01:00 level=INFO source=server.go:133 msg=mlx-runner msg="MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_8_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0"
time=2026-02-06T11:22:05.270+01:00 level=INFO source=server.go:133 msg=mlx-runner msg=" at /Users/runner/work/ollama/ollama/build/_deps/mlx-c-src/mlx/c/transforms.cpp:73"
[GIN] 2026/02/06 - 11:22:05 | 500 |  4.254860709s |       127.0.0.1 | POST     "/api/generate"

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.15.5

Originally created by @hwitzthum on GitHub (Feb 6, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14118 ### What is the issue? Config: macOS version (26.2) chip (M5) Ollama version (0.15.5) The MLX kernel error: MLX error: [metal::Device] Unable to load kernel. The model loads successfully into VRAM (11.9 GB), but when it tries to actually generate the image, it fails to load a Metal GPU kernel ### Relevant log output ```shell time=2026-02-06T11:22:05.172+01:00 level=INFO source=server.go:219 msg="mlx runner is ready" port=50052 time=2026-02-06T11:22:05.270+01:00 level=INFO source=server.go:133 msg=mlx-runner msg="MLX error: [metal::Device] Unable to load kernel affine_qmm_t_nax_bfloat16_t_gs_32_b_8_bm64_bn64_bk64_wm2_wn2_alN_true_batch_0" time=2026-02-06T11:22:05.270+01:00 level=INFO source=server.go:133 msg=mlx-runner msg=" at /Users/runner/work/ollama/ollama/build/_deps/mlx-c-src/mlx/c/transforms.cpp:73" [GIN] 2026/02/06 - 11:22:05 | 500 | 4.254860709s | 127.0.0.1 | POST "/api/generate" ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.15.5
GiteaMirror added the bug label 2026-04-22 19:03:56 -05:00
GiteaMirror commented 2026-04-22 19:03:58 -05:00
Author
Owner
Copy Link

@Extra-Citron-7630 commented on GitHub (Feb 7, 2026):

I think I am getting the same error

ollama --version
MLX: Failed to load symbol: mlx_metal_device_info
ollama version is 0.15.5
<!-- gh-comment-id:3865737706 --> @Extra-Citron-7630 commented on GitHub (Feb 7, 2026): I think I am getting the same error ``` ollama --version MLX: Failed to load symbol: mlx_metal_device_info ollama version is 0.15.5 ```
GiteaMirror commented 2026-04-22 19:03:58 -05:00
Author
Owner
Copy Link

@vvanmol commented on GitHub (Feb 8, 2026):

Facing the same issue on Apple M4 Max.

Relevant logs:

source=server.go:147 msg="starting mlx runner subprocess" exe=/opt/homebrew/Cellar/ollama/0.15.5/bin/ollama model=x/flux2-klein:latest port=49495 mode=imagegen
source=server.go:140 msg=mlx-runner msg="MLX: Failed to load symbol: mlx_metal_device_info"
source=server.go:140 msg=mlx-runner msg="MLX: Failed to load symbol: mlx_metal_device_info"
source=server.go:140 msg=mlx-runner msg="level=ERROR msg=\"unable to initialize MLX\" error=\"failed to load MLX function symbols\""
source=server.go:140 msg=mlx-runner msg="Error: failed to load MLX function symbols"
source=server.go:362 msg="stopping mlx runner subprocess" pid=1454

EDIT:

Brew dependencies are installed and up to date:

mlx-c ✔: stable 0.5.0 (bottled)
mlx ✔: stable 0.30.5 (bottled), HEAD
<!-- gh-comment-id:3866847921 --> @vvanmol commented on GitHub (Feb 8, 2026): Facing the same issue on Apple M4 Max. Relevant logs: ``` source=server.go:147 msg="starting mlx runner subprocess" exe=/opt/homebrew/Cellar/ollama/0.15.5/bin/ollama model=x/flux2-klein:latest port=49495 mode=imagegen source=server.go:140 msg=mlx-runner msg="MLX: Failed to load symbol: mlx_metal_device_info" source=server.go:140 msg=mlx-runner msg="MLX: Failed to load symbol: mlx_metal_device_info" source=server.go:140 msg=mlx-runner msg="level=ERROR msg=\"unable to initialize MLX\" error=\"failed to load MLX function symbols\"" source=server.go:140 msg=mlx-runner msg="Error: failed to load MLX function symbols" source=server.go:362 msg="stopping mlx runner subprocess" pid=1454 ``` EDIT: Brew dependencies are installed and up to date: ``` mlx-c ✔: stable 0.5.0 (bottled) mlx ✔: stable 0.30.5 (bottled), HEAD ```
GiteaMirror commented 2026-04-22 19:03:59 -05:00
Author
Owner
Copy Link

@Abioy commented on GitHub (Feb 9, 2026):

mlx-c 0.5.0 remove mlx_metal_device_info(). But ollama loads almost all dynamic symbols in libmlxc and fail with InitMLX even though most of them are not used in ollama.

workaround could be downgrade mlx-c to 0.4.1

I think the best way to fix this is remove those unused libmlxc apis in ollama.

<!-- gh-comment-id:3871919890 --> @Abioy commented on GitHub (Feb 9, 2026): mlx-c 0.5.0 remove mlx_metal_device_info(). But ollama loads almost all dynamic symbols in libmlxc and fail with InitMLX even though most of them are not used in ollama. workaround could be downgrade mlx-c to 0.4.1 I think the best way to fix this is remove those unused libmlxc apis in ollama.
GiteaMirror commented 2026-04-22 19:03:59 -05:00
Author
Owner
Copy Link

@chairbender commented on GitHub (Feb 9, 2026):

Just ran into this myself. It appears to be only an issue when installing via brew. If I had to speculate (not very familiar with brew myself), the latest version of mlx-c is 0.5.0 and that's the one the brew formula is installing. I'm not sure if brew is just unable to pin a transitive dep to a specific version or if the formula is somehow mis-configured.
Or, perhaps I ran brew upgrade after installing and it upgraded mlx-c as a side-effect.

I uninstalled viaa

# note it seems the ollama models are stored in ~/.ollama/models and AREN'T deleted by this so
# you won't have to re-download
brew uninstall --zap ollama

And then downloading the provided DMG linked from the main readme (and running it once). After that it's available on the command line and no more MLX errors.

<!-- gh-comment-id:3873663316 --> @chairbender commented on GitHub (Feb 9, 2026): Just ran into this myself. It appears to be only an issue when installing via brew. If I had to speculate (not very familiar with brew myself), the latest version of mlx-c is 0.5.0 and that's the one the brew formula is installing. I'm not sure if brew is just unable to pin a transitive dep to a specific version or if the formula is somehow mis-configured. Or, perhaps I ran brew upgrade after installing and it upgraded mlx-c as a side-effect. I uninstalled viaa ```bash # note it seems the ollama models are stored in ~/.ollama/models and AREN'T deleted by this so # you won't have to re-download brew uninstall --zap ollama ``` And then downloading the provided DMG linked from the main readme (and running it once). After that it's available on the command line and no more MLX errors.
GiteaMirror commented 2026-04-22 19:03:59 -05:00
Author
Owner
Copy Link

@kbrock commented on GitHub (Feb 10, 2026):

I can verify that downgrading from 5.0 to 4.1.1 fixes the problem.

Please don't script kiddie this.

If you don't know what I'm typing or how homebrew works, I really suggest using the DMG.
I'm including this because this group tends to do some pretty crazy stuff. So I assume you are advanced users...

ollama --version

curl -LO wget https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb
brew unlink mlx-c
HOMEBREW_DEVELOPER=1 brew install --formula ./mlx-c.rb
brew pin mlx-c # <=== WARNING

brew info mlx-c
ollama --version

This says never upgrade it.

Remember to brew unpin mlx-c

UPDATE: fixed curl. thnx

<!-- gh-comment-id:3875272015 --> @kbrock commented on GitHub (Feb 10, 2026): I can verify that downgrading from `5.0` to `4.1.1` fixes the problem. ## Please don't script kiddie this. If you don't know what I'm typing or how homebrew works, I really suggest using the DMG. I'm including this because this group tends to do some pretty crazy stuff. So I assume you are advanced users... ```bash ollama --version curl -LO wget https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb brew unlink mlx-c HOMEBREW_DEVELOPER=1 brew install --formula ./mlx-c.rb brew pin mlx-c # <=== WARNING brew info mlx-c ollama --version ``` This says never upgrade it. ## Remember to `brew unpin mlx-c` UPDATE: fixed curl. thnx
GiteaMirror commented 2026-04-22 19:04:00 -05:00
Author
Owner
Copy Link

@Extra-Citron-7630 commented on GitHub (Feb 10, 2026):

I can verify that downgrading from 5.0 to 4.1.1 fixes the problem.

Please don't script kiddie this.

If you don't know what I'm typing or how homebrew works, I really suggest using the DMG. I'm including this because this group tends to do some pretty crazy stuff. So I assume you are advanced users...

ollama --version

curl -O wget https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb
brew unlink mlx-c
HOMEBREW_DEVELOPER=1 brew install --formula ./mlx-c.rb
brew pin mlx-c # <=== WARNING

brew info mlx-c
ollama --version
This says never upgrade it.

Remember to brew unpin mlx-c

curl -LO https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb

This should work, I think you combined curl and wget

<!-- gh-comment-id:3875569463 --> @Extra-Citron-7630 commented on GitHub (Feb 10, 2026): > I can verify that downgrading from `5.0` to `4.1.1` fixes the problem. > > ## Please don't script kiddie this. > > If you don't know what I'm typing or how homebrew works, I really suggest using the DMG. I'm including this because this group tends to do some pretty crazy stuff. So I assume you are advanced users... > > ollama --version > > curl -O wget https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb > brew unlink mlx-c > HOMEBREW_DEVELOPER=1 brew install --formula ./mlx-c.rb > brew pin mlx-c # <=== WARNING > > brew info mlx-c > ollama --version > This says never upgrade it. > > ## Remember to `brew unpin mlx-c` curl -LO https://raw.githubusercontent.com/Homebrew/homebrew-core/e6183090dd4753cc12a9c47eff11832b044a0702/Formula/m/mlx-c.rb This should work, I think you combined curl and wget
GiteaMirror commented 2026-04-22 19:04:01 -05:00
Author
Owner
Copy Link

@vvanmol commented on GitHub (Feb 10, 2026):

Indeed downgrading mlx-c to 0.4.1 does fix the MLX errors.

<!-- gh-comment-id:3879132918 --> @vvanmol commented on GitHub (Feb 10, 2026): Indeed downgrading mlx-c to 0.4.1 does fix the MLX errors.
GiteaMirror commented 2026-04-22 19:04:02 -05:00
Author
Owner
Copy Link

@michaellmonaghan commented on GitHub (Feb 11, 2026):

Downgrading mlx-c appears to have worked on my system as well. Took a different approach creating a versioned recipe for mlx-c@0.4.1 in a local repo.

<!-- gh-comment-id:3885223111 --> @michaellmonaghan commented on GitHub (Feb 11, 2026): Downgrading mlx-c appears to have worked on my system as well. Took a different approach creating a versioned recipe for mlx-c@0.4.1 in a local repo.
GiteaMirror commented 2026-04-22 19:04:04 -05:00
Author
Owner
Copy Link

@michaellmonaghan commented on GitHub (Feb 11, 2026):

Homebrew issue has been filed https://github.com/Homebrew/homebrew-core/issues/266704

<!-- gh-comment-id:3885289111 --> @michaellmonaghan commented on GitHub (Feb 11, 2026): Homebrew issue has been filed https://github.com/Homebrew/homebrew-core/issues/266704
GiteaMirror commented 2026-04-22 19:04:06 -05:00
Author
Owner
Copy Link

@senki commented on GitHub (Feb 13, 2026):

Homebrew installed ollama upgraded to 0.15.6 (cli version) solved the problem for me even with 0.5.0 mlx-c. Previously ollama 0.15.5 displayed the MLX: Failed to load symbol: mlx_metal_device_info error.

Note no 0.16.x is available in Homebrew yet.

<!-- gh-comment-id:3894881376 --> @senki commented on GitHub (Feb 13, 2026): Homebrew installed ollama upgraded to 0.15.6 (cli version) solved the problem for me even with 0.5.0 mlx-c. Previously ollama 0.15.5 displayed the `MLX: Failed to load symbol: mlx_metal_device_info` error. Note no 0.16.x is available in Homebrew yet.
GiteaMirror commented 2026-04-22 19:04:08 -05:00
Author
Owner
Copy Link

@aihua commented on GitHub (Feb 13, 2026):

  • Reproduce:
  1. Download ollama from release https://github.com/ollama/ollama/releases/download/v0.16.1/ollama-darwin.tgz
  2. tar zxvf ollama-darwin.tgz && ollama serve && ollama run x/z-image-turbo

  • The Error:
    failed to load model: 500 Internal Server Error: mlx runner failed: model.norm.weight (exit: exit status 1)

  • The Log

time=2026-02-13T12:55:02.902+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.902+08:00 level=INFO msg=\"MLX library initialized\""
time=2026-02-13T12:55:02.905+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.905+08:00 level=INFO msg=\"starting mlx runner\" model=x/z-image-turbo:latest port=50799 mode=imagegen"
time=2026-02-13T12:55:02.911+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.911+08:00 level=INFO msg=\"detected image model type\" type=ZImagePipeline"
time=2026-02-13T12:55:02.911+08:00 level=INFO source=server.go:134 msg=mlx-runner msg="Loading Z-Image model from manifest: x/z-image-turbo:latest..."
time=2026-02-13T12:55:03.134+08:00 level=INFO source=server.go:134 msg=mlx-runner msg="  Loading tokenizer... ✓"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="Error: failed to create server: failed to load image model: failed to load zimage model: text encoder: load module: LoadModule: missing weights:"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.embed_tokens.weight"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.q_proj: failed to load quantized weight model.layers.0.self_attn.q_proj: tensor \"model.layers.0.self_attn.q_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.k_proj: failed to load quantized weight model.layers.0.self_attn.k_proj: tensor \"model.layers.0.self_attn.k_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.v_proj: failed to load quantized weight model.layers.0.self_attn.v_proj: tensor \"model.layers.0.self_attn.v_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.o_proj: failed to load quantized weight model.layers.0.self_attn.o_proj: tensor \"model.layers.0.self_attn.o_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.q_norm.weight"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.self_attn.k_norm.weight"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.mlp.gate_proj: failed to load quantized weight model.layers.0.mlp.gate_proj: tensor \"model.layers.0.mlp.gate_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.mlp.up_proj: failed to load quantized weight model.layers.0.mlp.up_proj: tensor \"model.layers.0.mlp.up_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.mlp.down_proj: failed to load quantized weight model.layers.0.mlp.down_proj: tensor \"model.layers.0.mlp.down_proj.weight\" not found"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.input_layernorm.weight"
time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.0.post_attention_layernorm.weight"
...   ...   ...
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.q_proj: failed to load quantized weight model.layers.35.self_attn.q_proj: tensor \"model.layers.35.self_attn.q_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.k_proj: failed to load quantized weight model.layers.35.self_attn.k_proj: tensor \"model.layers.35.self_attn.k_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.v_proj: failed to load quantized weight model.layers.35.self_attn.v_proj: tensor \"model.layers.35.self_attn.v_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.o_proj: failed to load quantized weight model.layers.35.self_attn.o_proj: tensor \"model.layers.35.self_attn.o_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.q_norm.weight"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.self_attn.k_norm.weight"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.mlp.gate_proj: failed to load quantized weight model.layers.35.mlp.gate_proj: tensor \"model.layers.35.mlp.gate_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.mlp.up_proj: failed to load quantized weight model.layers.35.mlp.up_proj: tensor \"model.layers.35.mlp.up_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.mlp.down_proj: failed to load quantized weight model.layers.35.mlp.down_proj: tensor \"model.layers.35.mlp.down_proj.weight\" not found"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.input_layernorm.weight"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.layers.35.post_attention_layernorm.weight"
time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="  model.norm.weight"
time=2026-02-13T12:55:04.082+08:00 level=INFO source=server.go:134 msg=mlx-runner msg="  Loading text encoder... "
time=2026-02-13T12:55:04.082+08:00 level=INFO source=server.go:363 msg="stopping mlx runner subprocess" pid=3488
<!-- gh-comment-id:3894931208 --> @aihua commented on GitHub (Feb 13, 2026): - Reproduce: 1. Download ollama from release [https://github.com/ollama/ollama/releases/download/v0.16.1/ollama-darwin.tgz](https://github.com/ollama/ollama/releases/download/v0.16.1/ollama-darwin.tgz) 2. `tar zxvf ollama-darwin.tgz && ollama serve && ollama run x/z-image-turbo` - The Error: `failed to load model: 500 Internal Server Error: mlx runner failed: model.norm.weight (exit: exit status 1)` - The Log ``` time=2026-02-13T12:55:02.902+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.902+08:00 level=INFO msg=\"MLX library initialized\"" time=2026-02-13T12:55:02.905+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.905+08:00 level=INFO msg=\"starting mlx runner\" model=x/z-image-turbo:latest port=50799 mode=imagegen" time=2026-02-13T12:55:02.911+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="time=2026-02-13T12:55:02.911+08:00 level=INFO msg=\"detected image model type\" type=ZImagePipeline" time=2026-02-13T12:55:02.911+08:00 level=INFO source=server.go:134 msg=mlx-runner msg="Loading Z-Image model from manifest: x/z-image-turbo:latest..." time=2026-02-13T12:55:03.134+08:00 level=INFO source=server.go:134 msg=mlx-runner msg=" Loading tokenizer... ✓" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg="Error: failed to create server: failed to load image model: failed to load zimage model: text encoder: load module: LoadModule: missing weights:" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.embed_tokens.weight" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.q_proj: failed to load quantized weight model.layers.0.self_attn.q_proj: tensor \"model.layers.0.self_attn.q_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.k_proj: failed to load quantized weight model.layers.0.self_attn.k_proj: tensor \"model.layers.0.self_attn.k_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.v_proj: failed to load quantized weight model.layers.0.self_attn.v_proj: tensor \"model.layers.0.self_attn.v_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.o_proj: failed to load quantized weight model.layers.0.self_attn.o_proj: tensor \"model.layers.0.self_attn.o_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.q_norm.weight" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.self_attn.k_norm.weight" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.mlp.gate_proj: failed to load quantized weight model.layers.0.mlp.gate_proj: tensor \"model.layers.0.mlp.gate_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.mlp.up_proj: failed to load quantized weight model.layers.0.mlp.up_proj: tensor \"model.layers.0.mlp.up_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.mlp.down_proj: failed to load quantized weight model.layers.0.mlp.down_proj: tensor \"model.layers.0.mlp.down_proj.weight\" not found" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.input_layernorm.weight" time=2026-02-13T12:55:04.034+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.0.post_attention_layernorm.weight" ... ... ... time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.q_proj: failed to load quantized weight model.layers.35.self_attn.q_proj: tensor \"model.layers.35.self_attn.q_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.k_proj: failed to load quantized weight model.layers.35.self_attn.k_proj: tensor \"model.layers.35.self_attn.k_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.v_proj: failed to load quantized weight model.layers.35.self_attn.v_proj: tensor \"model.layers.35.self_attn.v_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.o_proj: failed to load quantized weight model.layers.35.self_attn.o_proj: tensor \"model.layers.35.self_attn.o_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.q_norm.weight" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.self_attn.k_norm.weight" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.mlp.gate_proj: failed to load quantized weight model.layers.35.mlp.gate_proj: tensor \"model.layers.35.mlp.gate_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.mlp.up_proj: failed to load quantized weight model.layers.35.mlp.up_proj: tensor \"model.layers.35.mlp.up_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.mlp.down_proj: failed to load quantized weight model.layers.35.mlp.down_proj: tensor \"model.layers.35.mlp.down_proj.weight\" not found" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.input_layernorm.weight" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.layers.35.post_attention_layernorm.weight" time=2026-02-13T12:55:04.036+08:00 level=WARN source=server.go:141 msg=mlx-runner msg=" model.norm.weight" time=2026-02-13T12:55:04.082+08:00 level=INFO source=server.go:134 msg=mlx-runner msg=" Loading text encoder... " time=2026-02-13T12:55:04.082+08:00 level=INFO source=server.go:363 msg="stopping mlx runner subprocess" pid=3488 ```
GiteaMirror commented 2026-04-22 19:04:09 -05:00
Author
Owner
Copy Link

@kbrock commented on GitHub (Feb 13, 2026):

@hwitzthum Is this working for you? Can we close this issue?

@michaellmonaghan Thanks. It has been >15 years since I set up a formula. Did you just follow standard instructions? https://docs.brew.sh/How-to-Create-and-Maintain-a-Tap Huh, ollama formula is in json and not ruby? wow. times change.

@aihua We're talking about Homebrew. Update your homebrew formula and see if that works. Find the homebrew formula in https://formulae.brew.sh/formula/ollama
Or if you are not on a mac, then create a different issue.

<!-- gh-comment-id:3898343067 --> @kbrock commented on GitHub (Feb 13, 2026): @hwitzthum Is this working for you? Can we close this issue? @michaellmonaghan Thanks. It has been >15 years since I set up a formula. Did you just follow standard instructions? https://docs.brew.sh/How-to-Create-and-Maintain-a-Tap Huh, ollama formula is in json and not ruby? wow. times change. @aihua We're talking about Homebrew. Update your homebrew formula and see if that works. Find the homebrew formula in https://formulae.brew.sh/formula/ollama Or if you are not on a mac, then create a different issue.
GiteaMirror commented 2026-04-22 19:04:10 -05:00
Author
Owner
Copy Link

@humbertowoody commented on GitHub (Feb 14, 2026):

Hi @kbrock, I just wanted to comment before the issue gets closed that I am still presenting this problem, I am running 0.15.6 via homebrew:

❯ ollama --version
ollama version is 0.15.6

Server final logs:

ggml_metal_init: error: failed to initialize the Metal library
ggml_backend_metal_device_init: error: failed to allocate context
llama_init_from_model: failed to initialize the context: failed to initialize Metal backend
panic: unable to create llama context

goroutine 43 [running]:
github.com/ollama/ollama/runner/llamarunner.(*Server).loadModel(0x14000418140, {{0x0, 0x0, 0x0}, 0x0, 0x0, 0x0, {0x0, 0x0, 0x0}, ...}, ...)
	github.com/ollama/ollama/runner/llamarunner/runner.go:849 +0x268
created by github.com/ollama/ollama/runner/llamarunner.(*Server).load in goroutine 40
	github.com/ollama/ollama/runner/llamarunner/runner.go:934 +0x680
time=2026-02-13T20:39:32.446-06:00 level=ERROR source=server.go:304 msg="llama runner terminated" error="exit status 2"
time=2026-02-13T20:39:32.560-06:00 level=INFO source=sched.go:490 msg="Load failed" model=/Users/humalcocer/.ollama/models/blobs/sha256-3c168af1dea0a414299c7d9077e100ac763370e5a98b3c53801a958a47f0a5db error="llama runner process has terminated: error:failed to allocate context\nllama_init_from_model: failed to initialize the context: failed to initialize Metal backend"

I have been solving it by using 0.14.3 from official release as concluded on https://github.com/ollama/ollama/issues/13867 and it actually works, but I am missing out on latest features. I can provide any more information/logs required. MacBook Pro M5, 32G RAM.m

Edit:

I have done this tests by using phi4-mini:latest locally. In every case it works on 0.14.3 via tarball release, it doesn't on 0.15.6 via homebrew. Tried also phi3:latest, llama3:latest, gemma3:270m. Tried wiping out and re-pulling them from each version and had the same results. Happy to provide more details if needed.

<!-- gh-comment-id:3900521817 --> @humbertowoody commented on GitHub (Feb 14, 2026): Hi @kbrock, I just wanted to comment before the issue gets closed that I am still presenting this problem, I am running 0.15.6 via homebrew: ```console ❯ ollama --version ollama version is 0.15.6 ``` Server final logs: ```txt ggml_metal_init: error: failed to initialize the Metal library ggml_backend_metal_device_init: error: failed to allocate context llama_init_from_model: failed to initialize the context: failed to initialize Metal backend panic: unable to create llama context goroutine 43 [running]: github.com/ollama/ollama/runner/llamarunner.(*Server).loadModel(0x14000418140, {{0x0, 0x0, 0x0}, 0x0, 0x0, 0x0, {0x0, 0x0, 0x0}, ...}, ...) github.com/ollama/ollama/runner/llamarunner/runner.go:849 +0x268 created by github.com/ollama/ollama/runner/llamarunner.(*Server).load in goroutine 40 github.com/ollama/ollama/runner/llamarunner/runner.go:934 +0x680 time=2026-02-13T20:39:32.446-06:00 level=ERROR source=server.go:304 msg="llama runner terminated" error="exit status 2" time=2026-02-13T20:39:32.560-06:00 level=INFO source=sched.go:490 msg="Load failed" model=/Users/humalcocer/.ollama/models/blobs/sha256-3c168af1dea0a414299c7d9077e100ac763370e5a98b3c53801a958a47f0a5db error="llama runner process has terminated: error:failed to allocate context\nllama_init_from_model: failed to initialize the context: failed to initialize Metal backend" ``` I have been solving it by using 0.14.3 from official release as concluded on https://github.com/ollama/ollama/issues/13867 and it actually works, but I am missing out on latest features. I can provide any more information/logs required. MacBook Pro M5, 32G RAM.m Edit: I have done this tests by using `phi4-mini:latest` locally. In every case it works on 0.14.3 via tarball release, it doesn't on 0.15.6 via homebrew. Tried also `phi3:latest`, `llama3:latest`, `gemma3:270m`. Tried wiping out and re-pulling them from each version and had the same results. Happy to provide more details if needed.
GiteaMirror commented 2026-04-22 19:04:11 -05:00
Author
Owner
Copy Link

@ghuh commented on GitHub (Feb 18, 2026):

For ollama v0.16.2 installed via brew, I'm getting

2026/02/18 11:17:55 WARN MLX dynamic library not available error="failed to load MLX dynamic library ...

Anyone else seeing this?

<!-- gh-comment-id:3922433052 --> @ghuh commented on GitHub (Feb 18, 2026): For ollama v0.16.2 installed via brew, I'm getting ``` 2026/02/18 11:17:55 WARN MLX dynamic library not available error="failed to load MLX dynamic library ... ``` Anyone else seeing this?
GiteaMirror commented 2026-04-22 19:04:11 -05:00
Author
Owner
Copy Link

@AbeEstrada commented on GitHub (Feb 18, 2026):

For ollama v0.16.2 installed via brew, I'm getting

2026/02/18 11:17:55 WARN MLX dynamic library not available error="failed to load MLX dynamic library ...

Anyone else seeing this?

Yes, I get the same warning after the update

<!-- gh-comment-id:3922526999 --> @AbeEstrada commented on GitHub (Feb 18, 2026): > For ollama v0.16.2 installed via brew, I'm getting > > ``` > 2026/02/18 11:17:55 WARN MLX dynamic library not available error="failed to load MLX dynamic library ... > ``` > > Anyone else seeing this? Yes, I get the same warning after the update
GiteaMirror commented 2026-04-22 19:04:12 -05:00
Author
Owner
Copy Link

@kbrock commented on GitHub (Feb 18, 2026):

ugh. thought they just fixed this.
Seems to happen every time mlx upgrades

<!-- gh-comment-id:3922878724 --> @kbrock commented on GitHub (Feb 18, 2026): ugh. thought they just fixed this. Seems to happen every time mlx upgrades
GiteaMirror commented 2026-04-22 19:04:12 -05:00
Author
Owner
Copy Link

@ghuh commented on GitHub (Feb 19, 2026):

Looks like there is a PR out to fix this

<!-- gh-comment-id:3928134303 --> @ghuh commented on GitHub (Feb 19, 2026): Looks like there is a [PR](https://github.com/ollama/ollama/pull/14322) out to fix this
GiteaMirror commented 2026-04-22 19:04:13 -05:00
Author
Owner
Copy Link

@hwitzthum commented on GitHub (Feb 19, 2026):

Great! All the attempts have not been stable enough. Best

Von: Kevin Hayen @.>
Datum: Donnerstag, 19. Februar 2026 um 16:50
An: ollama/ollama @.>
Cc: hwitzthum @.>, Mention @.>
Betreff: Re: [ollama/ollama] MLX Error (Issue #14118)

[https://avatars.githubusercontent.com/u/7354238?s=20&v=4]ghuh left a comment (ollama/ollama#14118)https://github.com/ollama/ollama/issues/14118#issuecomment-3928134303

Looks like there is a PRhttps://github.com/ollama/ollama/pull/14322 out to fix this

—
Reply to this email directly, view it on GitHubhttps://github.com/ollama/ollama/issues/14118#issuecomment-3928134303, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BFFJD4VXO5CH2MBYFJMPDPT4MXLVDAVCNFSM6AAAAACUGW6M36VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSMRYGEZTIMZQGM.
You are receiving this because you were mentioned.

<!-- gh-comment-id:3929472072 --> @hwitzthum commented on GitHub (Feb 19, 2026): Great! All the attempts have not been stable enough. Best Von: Kevin Hayen ***@***.***> Datum: Donnerstag, 19. Februar 2026 um 16:50 An: ollama/ollama ***@***.***> Cc: hwitzthum ***@***.***>, Mention ***@***.***> Betreff: Re: [ollama/ollama] MLX Error (Issue #14118) [https://avatars.githubusercontent.com/u/7354238?s=20&v=4]ghuh left a comment (ollama/ollama#14118)<https://github.com/ollama/ollama/issues/14118#issuecomment-3928134303> Looks like there is a PR<https://github.com/ollama/ollama/pull/14322> out to fix this — Reply to this email directly, view it on GitHub<https://github.com/ollama/ollama/issues/14118#issuecomment-3928134303>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BFFJD4VXO5CH2MBYFJMPDPT4MXLVDAVCNFSM6AAAAACUGW6M36VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSMRYGEZTIMZQGM>. You are receiving this because you were mentioned.
GiteaMirror commented 2026-04-22 19:04:13 -05:00
Author
Owner
Copy Link

@jloh888 commented on GitHub (Feb 20, 2026):

A workaround for this, which I think this is a problem with the way homebrew handles dependencies, is to just go to the download section of ollama and d/l it from there.
Do a 'brew uninstall ollama' first.

<!-- gh-comment-id:3934807551 --> @jloh888 commented on GitHub (Feb 20, 2026): A workaround for this, which I think this is a problem with the way homebrew handles dependencies, is to just go to the download section of ollama and d/l it from there. Do a 'brew uninstall ollama' first.
GiteaMirror commented 2026-04-22 19:04:14 -05:00
Author
Owner
Copy Link

@LouGrossi commented on GitHub (Feb 20, 2026):

Fix: "MLX dynamic library not available" (Homebrew Ollama on macOS)

When you run ollama you may see:

WARN MLX dynamic library not available error="failed to load MLX dynamic library (searched: [path1 path2])"

Ollama only looks in the directories listed in that searched: line. The Homebrew MLX libs live elsewhere, so it never finds them. Symlinking those libs into one of the searched paths fixes it.

Step 1: Get the exact search paths

Run any Ollama command and copy the full error line (including the searched: part):

ollama list 2>&1

Example:

WARN MLX dynamic library not available error="failed to load MLX dynamic library (searched: [/opt/homebrew/Cellar/ollama/0.16.2/bin /Users/YOUR_USERNAME/build/lib/ollama])"
NAME    ID    SIZE    MODIFIED

From that line, note the two paths inside searched: [...]:

  • First path: something like /opt/homebrew/Cellar/ollama/<version>/bin
  • Second path: something like /Users/<your_username>/build/lib/ollama

You'll symlink the libs into one of these (Step 3).

Step 2: Find where the MLX libraries are

Homebrew put them under its prefix. Get the paths:

# MLX (Apple)
ls "$(brew --prefix mlx)/lib/libmlx.dylib"

# MLX C API (used by Ollama)
ls "$(brew --prefix mlx-c)/lib/libmlxc.dylib"

If both commands print a path, you're good. Typical values:

  • $(brew --prefix mlx)/lib → e.g. /opt/homebrew/opt/mlx/lib
  • $(brew --prefix mlx-c)/lib → e.g. /opt/homebrew/opt/mlx-c/lib

Step 3: Choose the directory to symlink into

Use one of the two paths from Step 1:

Option Path (from error) Who can write
A Second path: .../build/lib/ollama (under your home) You (no sudo)
B First path: .../Cellar/ollama/.../bin Needs sudo

Recommended: Use the second path (the one under your home, e.g. /Users/<you>/build/lib/ollama). Then you don't need sudo and the fix is per user.

Set a variable to that directory (replace with the real path from your error):

# Example: use the path under your home (second path from the error)
OLLAMA_LIB_SEARCH_DIR="$HOME/build/lib/ollama"

If you chose the Cellar path (Option B), use that path instead and run the next steps with sudo where needed.

Step 4: Create the directory and symlink the libraries

# Create the directory Ollama searches
mkdir -p "$OLLAMA_LIB_SEARCH_DIR"

# Symlink the MLX libraries into it
ln -sf "$(brew --prefix mlx)/lib/libmlx.dylib"    "$OLLAMA_LIB_SEARCH_DIR/"
ln -sf "$(brew --prefix mlx-c)/lib/libmlxc.dylib" "$OLLAMA_LIB_SEARCH_DIR/"

If you chose the Cellar path and don't have write access, use sudo for mkdir and ln and set OLLAMA_LIB_SEARCH_DIR to that path (e.g. /opt/homebrew/Cellar/ollama/0.16.2/bin).

Step 5: Confirm the fix

Run Ollama again; the MLX warning should be gone:

ollama list

You should see only the model table (and no "MLX dynamic library not available" line).

One-liner (after you know the search path)

If your error lists $HOME/build/lib/ollama as one of the search paths, you can do everything in one go:

mkdir -p "$HOME/build/lib/ollama" && \
ln -sf "$(brew --prefix mlx)/lib/libmlx.dylib" "$HOME/build/lib/ollama/" && \
ln -sf "$(brew --prefix mlx-c)/lib/libmlxc.dylib" "$HOME/build/lib/ollama/"

Then run ollama list again to verify.

Why this works

  • The error message tells you exactly which directories Ollama searches for the MLX library (searched: [path1 path2]).
  • Homebrew installs the real libraries under brew --prefix mlx and brew --prefix mlx-c.
  • By creating one of the searched directories and putting symlinks to those libs there, Ollama finds them and the warning goes away.
<!-- gh-comment-id:3937573121 --> @LouGrossi commented on GitHub (Feb 20, 2026): # Fix: "MLX dynamic library not available" (Homebrew Ollama on macOS) When you run `ollama` you may see: ``` WARN MLX dynamic library not available error="failed to load MLX dynamic library (searched: [path1 path2])" ``` Ollama only looks in the directories listed in that `searched:` line. The Homebrew MLX libs live elsewhere, so it never finds them. Symlinking those libs into one of the searched paths fixes it. --- ## Step 1: Get the exact search paths Run any Ollama command and copy the full error line (including the `searched:` part): ```bash ollama list 2>&1 ``` Example: ``` WARN MLX dynamic library not available error="failed to load MLX dynamic library (searched: [/opt/homebrew/Cellar/ollama/0.16.2/bin /Users/YOUR_USERNAME/build/lib/ollama])" NAME ID SIZE MODIFIED ``` From that line, note the **two paths** inside `searched: [...]`: - **First path:** something like `/opt/homebrew/Cellar/ollama/<version>/bin` - **Second path:** something like `/Users/<your_username>/build/lib/ollama` You'll symlink the libs into **one** of these (Step 3). --- ## Step 2: Find where the MLX libraries are Homebrew put them under its prefix. Get the paths: ```bash # MLX (Apple) ls "$(brew --prefix mlx)/lib/libmlx.dylib" # MLX C API (used by Ollama) ls "$(brew --prefix mlx-c)/lib/libmlxc.dylib" ``` If both commands print a path, you're good. Typical values: - `$(brew --prefix mlx)/lib` → e.g. `/opt/homebrew/opt/mlx/lib` - `$(brew --prefix mlx-c)/lib` → e.g. `/opt/homebrew/opt/mlx-c/lib` --- ## Step 3: Choose the directory to symlink into Use **one** of the two paths from Step 1: | Option | Path (from error) | Who can write | |--------|-------------------|----------------| | **A** | Second path: `.../build/lib/ollama` (under your home) | You (no sudo) | | **B** | First path: `.../Cellar/ollama/.../bin` | Needs `sudo` | **Recommended:** Use the **second** path (the one under your home, e.g. `/Users/<you>/build/lib/ollama`). Then you don't need sudo and the fix is per user. Set a variable to that directory (replace with the real path from your error): ```bash # Example: use the path under your home (second path from the error) OLLAMA_LIB_SEARCH_DIR="$HOME/build/lib/ollama" ``` If you chose the Cellar path (Option B), use that path instead and run the next steps with `sudo` where needed. --- ## Step 4: Create the directory and symlink the libraries ```bash # Create the directory Ollama searches mkdir -p "$OLLAMA_LIB_SEARCH_DIR" # Symlink the MLX libraries into it ln -sf "$(brew --prefix mlx)/lib/libmlx.dylib" "$OLLAMA_LIB_SEARCH_DIR/" ln -sf "$(brew --prefix mlx-c)/lib/libmlxc.dylib" "$OLLAMA_LIB_SEARCH_DIR/" ``` If you chose the Cellar path and don't have write access, use `sudo` for `mkdir` and `ln` and set `OLLAMA_LIB_SEARCH_DIR` to that path (e.g. `/opt/homebrew/Cellar/ollama/0.16.2/bin`). --- ## Step 5: Confirm the fix Run Ollama again; the MLX warning should be gone: ```bash ollama list ``` You should see only the model table (and no "MLX dynamic library not available" line). --- ## One-liner (after you know the search path) If your error lists `$HOME/build/lib/ollama` as one of the search paths, you can do everything in one go: ```bash mkdir -p "$HOME/build/lib/ollama" && \ ln -sf "$(brew --prefix mlx)/lib/libmlx.dylib" "$HOME/build/lib/ollama/" && \ ln -sf "$(brew --prefix mlx-c)/lib/libmlxc.dylib" "$HOME/build/lib/ollama/" ``` Then run `ollama list` again to verify. --- ## Why this works - The **error message tells you exactly which directories** Ollama searches for the MLX library (`searched: [path1 path2]`). - Homebrew installs the real libraries under `brew --prefix mlx` and `brew --prefix mlx-c`. - By creating one of the searched directories and putting symlinks to those libs there, Ollama finds them and the warning goes away.
GiteaMirror commented 2026-04-22 19:04:14 -05:00
Author
Owner
Copy Link

@dvessel commented on GitHub (Feb 21, 2026):

Recent update through homebrew still outputs the error (0.16.3). A temp fix is to just soft link the dylib into the ollama bin directory. Next time it updates, the link will be discarded and hopefully it gets fixed.

ln -s `brew --prefix mlx-c`/lib/libmlxc.dylib `brew --prefix ollama`/bin
<!-- gh-comment-id:3938612368 --> @dvessel commented on GitHub (Feb 21, 2026): Recent update through homebrew still outputs the error (0.16.3). A temp fix is to just soft link the dylib into the ollama bin directory. Next time it updates, the link will be discarded *and hopefully it gets fixed*. ```sh ln -s `brew --prefix mlx-c`/lib/libmlxc.dylib `brew --prefix ollama`/bin ```
GiteaMirror commented 2026-04-22 19:04:14 -05:00
Author
Owner
Copy Link

@Mottl commented on GitHub (Feb 21, 2026):

Another option is to add OLLAMA_LIBRARY_PATH to environment vars.
For bash:
export OLLAMA_LIBRARY_PATH=$(brew --prefix)/lib
For fish:
set -Ux OLLAMA_LIBRARY_PATH $(brew --prefix)/lib

<!-- gh-comment-id:3938763662 --> @Mottl commented on GitHub (Feb 21, 2026): Another option is to add `OLLAMA_LIBRARY_PATH` to environment vars. For bash: `export OLLAMA_LIBRARY_PATH=$(brew --prefix)/lib` For fish: `set -Ux OLLAMA_LIBRARY_PATH $(brew --prefix)/lib`
GiteaMirror commented 2026-04-22 19:04:15 -05:00
Author
Owner
Copy Link

@kbrock commented on GitHub (Feb 22, 2026):

@LouGrossi Nice write up. For me, the path A is actually using $PWD and not $HOME. Think that worked for you b/c you were in ~ when you did your research?

@Mottl - Good find. Think the issue is that we set the mlx path via rpath() rather than an absolute path. Since mlx-c is linked into $(brew --path)/lib, we should set that into the formula. Then we can avoid having the end user set the OLLAMA_LIBRARY_PATH.

Looking into the formula. Someone please ping if you already solved this.

<!-- gh-comment-id:3939881435 --> @kbrock commented on GitHub (Feb 22, 2026): @LouGrossi Nice write up. For me, the path A is actually using `$PWD` and not `$HOME`. Think that worked for you b/c you were in ~ when you did your research? @Mottl - Good find. Think the issue is that we set the mlx path via `rpath()` rather than an absolute path. Since `mlx-c` is linked into `$(brew --path)/lib`, we should set that into the formula. Then we can avoid having the end user set the `OLLAMA_LIBRARY_PATH`. Looking into the formula. Someone please ping if you already solved this.
GiteaMirror commented 2026-04-22 19:04:16 -05:00
Author
Owner
Copy Link

@ilaikim99 commented on GitHub (Feb 28, 2026):

This is a transitive dependency version conflict — mlx-c 0.5.0 removed a symbol that Ollama loads at startup. Three ways to fix it: https://cacheoverflow.dev/blog/fJkb7kkq

<!-- gh-comment-id:3977283830 --> @ilaikim99 commented on GitHub (Feb 28, 2026): This is a transitive dependency version conflict — mlx-c 0.5.0 removed a symbol that Ollama loads at startup. Three ways to fix it: https://cacheoverflow.dev/blog/fJkb7kkq
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#34971
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?