github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-18 06:06:18 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #15623] gemma4:26b-nvfp4 has wrong metadata #56480

New Issue
Open
opened 2026-04-29 10:53:03 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-29 10:53:03 -05:00
Owner
Copy Link

Originally created by @empperi on GitHub (Apr 16, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15623

What is the issue?

Metadata on gemma4:26b-nvfp4 seems to be incorrect and it is marked to require MacOS. This must be a mistake since nvfp4 is a Nvidia Blackwell specific feature and no Apple product ships with that. One has to use external GPUs to get Nvidia into Apple world.

Relevant log output

❯ ollama run gemma4:26b-nvfp4
pulling manifest
Error: pull model manifest: 412: this model requires macOS

OS

Windows

GPU

Nvidia

CPU

AMD

Ollama version

0.20.7

Originally created by @empperi on GitHub (Apr 16, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15623 ### What is the issue? Metadata on `gemma4:26b-nvfp4` seems to be incorrect and it is marked to require MacOS. This must be a mistake since nvfp4 is a Nvidia Blackwell specific feature and no Apple product ships with that. One has to use external GPUs to get Nvidia into Apple world. ### Relevant log output ```shell ❯ ollama run gemma4:26b-nvfp4 pulling manifest Error: pull model manifest: 412: this model requires macOS ``` ### OS Windows ### GPU Nvidia ### CPU AMD ### Ollama version 0.20.7
GiteaMirror added the bug label 2026-04-29 10:53:03 -05:00
GiteaMirror commented 2026-04-29 10:53:06 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 16, 2026):

nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model.

<!-- gh-comment-id:4261295287 --> @rick-github commented on GitHub (Apr 16, 2026): nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model.
GiteaMirror commented 2026-04-29 10:53:07 -05:00
Author
Owner
Copy Link

@empperi commented on GitHub (Apr 16, 2026):

nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model.

OK, but NVFP4 is Nvidia specific technology created for Blackwell. Even if MLX engine has included support for format it makes no sense to restrict that to only MacOS unless NVFP4 support in Ollama is not included for Blackwell which would be weird.

Since you are a collaborator I guess you can clarify this? And if it truly is like that then why? No development resources? I can appreciate hard work of opensource community but as it is now (if true on only being supported on MLX) it is hugely confusing.

<!-- gh-comment-id:4261365834 --> @empperi commented on GitHub (Apr 16, 2026): > nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model. OK, but NVFP4 is Nvidia specific technology created for Blackwell. Even if MLX engine has included support for format it makes no sense to restrict that to only MacOS unless NVFP4 support in Ollama is not included for Blackwell which would be weird. Since you are a collaborator I guess you can clarify this? And if it truly is like that then why? No development resources? I can appreciate hard work of opensource community but as it is now (if true on only being supported on MLX) it is hugely confusing.
GiteaMirror commented 2026-04-29 10:53:07 -05:00
Author
Owner
Copy Link

@SingularityMan commented on GitHub (Apr 16, 2026):

nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model.

OK, but NVFP4 is Nvidia specific technology created for Blackwell. Even if MLX engine has included support for format it makes no sense to restrict that to only MacOS unless NVFP4 support in Ollama is not included for Blackwell which would be weird.

Since you are a collaborator I guess you can clarify this? And if it truly is like that then why? No development resources? I can appreciate hard work of opensource community but as it is now (if true on only being supported on MLX) it is hugely confusing.

Yup, Blackwell is supposed to run this, not just MLX.

<!-- gh-comment-id:4261407206 --> @SingularityMan commented on GitHub (Apr 16, 2026): > > nvfp4 is supported by the MLX engine. The MLX engine is currently only supported on macOS. So macOS is required to run the model. > > OK, but NVFP4 is Nvidia specific technology created for Blackwell. Even if MLX engine has included support for format it makes no sense to restrict that to only MacOS unless NVFP4 support in Ollama is not included for Blackwell which would be weird. > > Since you are a collaborator I guess you can clarify this? And if it truly is like that then why? No development resources? I can appreciate hard work of opensource community but as it is now (if true on only being supported on MLX) it is hugely confusing. Yup, Blackwell is supposed to run this, not just MLX.
GiteaMirror commented 2026-04-29 10:53:08 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 16, 2026):

It's currently only available on macOS because that's the platform that MLX engine currently runs on. Work is in progress to make the MLX engine available on Linux and Windows. Support for nvfp4 may come to other backends in time, but at the moment, ir's MLX only and hence macOS only.

$ for i in qwen3.6:35b-a3b-bf16 qwen3.6:35b-a3b-mlx-bf16 ; do echo $i ; ollama run $i 'why is the sky blue?' --verbose 2>&1 | grep ^eval.rate ; ollama ps ; done
qwen3.6:35b-a3b-bf16
eval rate:            105.47 tokens/s
qwen3.6:35b-a3b-mlx-bf16
eval rate:            126.98 tokens/s

NAME                        ID              SIZE     PROCESSOR    CONTEXT    UNTIL   
qwen3.6:35b-a3b-bf16        94061ddd23a7    82 GB    100% GPU     262144     Forever    
qwen3.6:35b-a3b-mlx-bf16    6d7a4cae9475    69 GB    100% GPU     262144     Forever
<!-- gh-comment-id:4262043358 --> @rick-github commented on GitHub (Apr 16, 2026): It's currently only available on macOS because that's the platform that MLX engine currently runs on. Work is in progress to make the MLX engine available on Linux and Windows. Support for nvfp4 may come to other backends in time, but at the moment, ir's MLX only and hence macOS only. ```console $ for i in qwen3.6:35b-a3b-bf16 qwen3.6:35b-a3b-mlx-bf16 ; do echo $i ; ollama run $i 'why is the sky blue?' --verbose 2>&1 | grep ^eval.rate ; ollama ps ; done qwen3.6:35b-a3b-bf16 eval rate: 105.47 tokens/s qwen3.6:35b-a3b-mlx-bf16 eval rate: 126.98 tokens/s ``` ``` NAME ID SIZE PROCESSOR CONTEXT UNTIL qwen3.6:35b-a3b-bf16 94061ddd23a7 82 GB 100% GPU 262144 Forever qwen3.6:35b-a3b-mlx-bf16 6d7a4cae9475 69 GB 100% GPU 262144 Forever ```
GiteaMirror commented 2026-04-29 10:53:09 -05:00
Author
Owner
Copy Link

@Mati0kez commented on GitHub (Apr 16, 2026):

Agreed — nvfp4 is Nvidia Blackwell-specific and no Apple product ships with that. This looks like an incorrect platform/magic number in the model's metadata file. Happy to fix the metadata.

<!-- gh-comment-id:4262397057 --> @Mati0kez commented on GitHub (Apr 16, 2026): Agreed — nvfp4 is Nvidia Blackwell-specific and no Apple product ships with that. This looks like an incorrect platform/magic number in the model's metadata file. Happy to fix the metadata.
GiteaMirror commented 2026-04-29 10:53:10 -05:00
Author
Owner
Copy Link

@Mati0kez commented on GitHub (Apr 16, 2026):

Agreed — nvfp4 is Nvidia Blackwell-specific and no Apple product ships with that. This looks like incorrect platform/magic number in the model's metadata. Happy to fix the metadata.

<!-- gh-comment-id:4262435683 --> @Mati0kez commented on GitHub (Apr 16, 2026): Agreed — nvfp4 is Nvidia Blackwell-specific and no Apple product ships with that. This looks like incorrect platform/magic number in the model's metadata. Happy to fix the metadata.
GiteaMirror commented 2026-04-29 10:53:11 -05:00
Author
Owner
Copy Link

@danieltomasz commented on GitHub (Apr 16, 2026):

@Mati0kez please read the official announcement, nvfp4 initial support was announced by the end of March, this is qwen, but also nvfpr : https://ollama.com/blog/mlx

Please make sure you have a Mac with more than 32GB of unified memory.

https://forums.developer.nvidia.com/t/ollama-with-nvfp4-support/365286

<!-- gh-comment-id:4263525737 --> @danieltomasz commented on GitHub (Apr 16, 2026): @Mati0kez please read the official announcement, nvfp4 initial support was announced by the end of March, this is qwen, but also nvfpr : https://ollama.com/blog/mlx >Please make sure you have a Mac with more than 32GB of unified memory. https://forums.developer.nvidia.com/t/ollama-with-nvfp4-support/365286
GiteaMirror commented 2026-04-29 10:53:11 -05:00
Author
Owner
Copy Link

@PureBlissAK commented on GitHub (Apr 18, 2026):

🤖 Automated Triage & Analysis Report

Issue: #15623
Analyzed: 2026-04-18T18:19:22.613763

Analysis

  • Type: unknown
  • Severity: medium
  • Components: unknown

Implementation Plan

  • Effort: medium
  • Steps:

This issue has been triaged and marked for implementation.

<!-- gh-comment-id:4274304769 --> @PureBlissAK commented on GitHub (Apr 18, 2026): <!-- ollama-issue-orchestrator:v1 issue:15623 --> ## 🤖 Automated Triage & Analysis Report **Issue**: #15623 **Analyzed**: 2026-04-18T18:19:22.613763 ### Analysis - **Type**: unknown - **Severity**: medium - **Components**: unknown ### Implementation Plan - **Effort**: medium - **Steps**: *This issue has been triaged and marked for implementation.*
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#56480
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?