GGUF models doesnt'work anymore

[thodem]

What is the issue?

Since Ollama 0.30, some models that previously worked on Ollama 0.24 have stopped functioning. These models include:

• Gemma4:26b-a4b-it-q8_0
• Deepseek-php33-mlx:latest
• Deepseek-coder-v2:16b
• Mistral-small3.2:24b

However, these models still work:

• Qwen3.6:35b-a3b-coding-mxfp8
• Gemma4:26b-mlx

I'm not sure, if this is a homebrew issue or local on my configuration.
Hardware: Macbook pro, Apple M4 Max 64GB, macOS-26.4.1-arm64-arm-64bit-Mach-O,
ollama installed with homebrew.

Relevant log output

failed to create server" model=deepseek-php33-mlx:latest error="error starting llama-server: llama-server binary not found (checked: /opt/homebrew/Cellar/ollama/0.30.0/libexec/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/build/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin-arm64/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin_arm64/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin/llama-server, /opt/homebrew/var/build/lib/ollama/llama-server, /opt/homebrew/var/dist/darwin-arm64/lib/ollama/llama-server, /opt/homebrew/var/dist/darwin_arm64/lib/ollama/llama-server, /opt/homebrew/var/dist/darwin/llama-server). Run 'cmake -S llama/server --preset cpu && cmake --build --preset cpu' first "

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.30.0

[Yuhao-Zhuang-gif]

I met the similar issue for creating models including unsloth version of

• Qwen3.6-27B-Q4_K_M
• gemma-4-26B-A4B-it-UD-Q8_K_XL

OS: Linux, GPU Nvidia, CPU Intel,

Relevant log:

xxx:~/.models/Qwen$ ollama create qwen36_27B -f Modelfile_Qwen3.6_27B
gathering model components
copying file sha256:5ed60d0af4650a854b1755bd392f9aef4872643dc25a254bc68043fa638392a0 100%
parsing GGUF
Error: failed to validate GGUF with llama-quantize without compatibility patches: llama-quantize failed: exit status 1

[oshytiko]

Same for me!

All MLX models work, but any GGUF will give something like:

Error: 500 Internal Server Error: error starting llama-server: llama-server binary not found (checked: /opt/homebrew/Cellar/ollama/0.30.0/libexec/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/build/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin-arm64/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin_arm64/lib/ollama/llama-server, /opt/homebrew/Cellar/ollama/0.30.0/libexec/dist/darwin/llama-server, /Users/OSHYTIKO/build/lib/ollama/llama-server, /Users/OSHYTIKO/dist/darwin-arm64/lib/ollama/llama-server, /Users/OSHYTIKO/dist/darwin_arm64/lib/ollama/llama-server, /Users/OSHYTIKO/dist/darwin/llama-server). Run 'cmake -S llama/server --preset cpu && cmake --build --preset cpu' first 

[thodem]

As far as I understand it, Ollama 0.30 requires a separate llama-server for the GGUF models. At least with Homebrew, this isn't currently installed automatically.
In any case, the following command:
find /opt/homebrew -name llama-server returns no output, and reinstalling Ollama after uninstalling and cleaning up didn’t change anything here either. The llama-server isn’t found, but it also wasn’t installed.
It would be interesting to know whether Illama Server is installed when using the official .dmg package. I just don't have time to check right now.

[oshytiko]

I was installing llama.cpp separately, and the error was the same. Strangely enough...

[rick-github]

The brew recipe needs to be updated.

[oshytiko]

I don't think so, the recipe change for formula seems to be trivial Homebrew/homebrew-core@17c0ff2

But what I can say also, I have tried to installing ollama-app, and it works correctly.

My best guess is that the bug was introduced in the source code of the latest upstream release. The Ollama 0.3.0 release notes explicitly mention integrating llama.cpp to augment the MLX engine on Apple Silicon, which seems like the most likely area where this CLI regression was introduced.:

[oshytiko]

And also issue seems to be reproducible on Linux... yes, brew is also available there, but most likely the executable was built from source, so something is fishy here...

[thenumber8equalsign]

Same problem here, I use the archlinux package version and it errors, I downgrade to 0.24.0 and it works fine, I install llama.cpp-vulkan from the aur and it works. I checked systemctl status ollama while a model was running on version 0.24.0 and it seems to open ollama runner --ollama-engine --model [some other stuff]. Now with version 0.30.0 it seems to try to run llama-server from $PATH, but since it doesn't exist it errors.

[rick-github]

Again, the package build needs to be updated.

[xzeyo]

same issue for arch

running a headless arch VM with rtx 3060 passed to it, after updating to 0.30 i am having similar issue loading models

ollama[754]: time=2026-06-02T22:38:00.225+05:30 level=INFO source=sched.go:591 msg="failed to create server" model=gemma4:e4b error="error starting llama-server: llama-server binary not found (checked: /usr/bin/llama-server, /usr/lib/ollama/llama-server, /usr/bin/build/lib/ollama/llama-server, /usr/bin/dist/linux-amd64/lib/ollama/llama-server, /usr/bin/dist/linux_amd64/lib/ollama/llama-server, /build/lib/ollama/llama-server, /dist/linux-amd64/lib/ollama/llama-server, /dist/linux_amd64/lib/ollama/llama-server). Run 'cmake -S llama/server --preset cpu && cmake --build --preset cpu' first "