Error: [ollama run qwen3.6:27b-coding-nvfp4] pull model manifest: 412: this model requires macOS

[master-genius]

What is the issue?

$ ollama run qwen3.6:27b-coding-nvfp4

Error: pull model manifest: 412: this model requires macOS

Relevant log output

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.21.3-rc0

[princekin0820-art]

Ollama use MLX in MacOS taht support nvfp4. In the rest, ollama use llama.cpp that not yet support nvfp4 in current version.

[Thomas-git]

llama.cpp now supports nvfp4 with blackwell GPU. Is ollama planing to use it anytime soon ?

[vt-alt]

I also wonder what are these -coding- models are, because on Qwen blog announcement they are not mentioned. https://qwen.ai/blog?id=qwen3.6-27b From the Model table we see they don't support Image input, but what these models are, where they are from, are these Ollama's team fine-tunes?

[rick-github]

The coding models have different parameters (temperature, etc). They don't support images because the MLX backend they run on is not yet vision capable. They are otherwise unmodified.

[paulpall]

Why was this closed? Is it a duplicate of #16056?

[rick-github]

Answered here: #15854 (comment)

[vt-alt]

The coding models have different parameters (temperature, etc). They don't support images because the MLX backend they run on is not yet vision capable. They are otherwise unmodified.

@rick-github That's useful to know! Thanks! I hope it's explained in the README.

[paulpall]

Answered here: #15854 (comment)

Could you elaborate on how that's tied to the issue Error: pull model manifest: 412: this model requires macOS because I was under the impression this was due to nvfp4 support? Has there been a fix that I've missed?

[rick-github]

The OP wants to run qwen3.6:27b-coding-nvfp4. NVFP4 support is via the MLX engine. The MLX engine currently requires macOS. Transitively, running qwen3.6:27b-coding-nvfp4 requires macOS.