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 #4600] Fast-copy files on ollama create if accessible #49399

New Issue
Open
opened 2026-04-28 11:40:36 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-28 11:40:36 -05:00
Owner
Copy Link

Originally created by @spott on GitHub (May 24, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4600

ollama create given a model file which references a local .gguf file currently does a regular copy of the .gguf file.

A copy on write would save disk space and be faster, allowing people to have gguf files in multiple places without paying the disk space penalty.

This is only possible on a few different file systems. APFS is one (on all modern Macs), Btrfs, zfs, etc.

Changing cp to cp -c should do it on Macs. cp --reflink=auto should work on linux.

Originally created by @spott on GitHub (May 24, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4600 `ollama create` given a model file which references a local .gguf file currently does a regular copy of the .gguf file. A copy on write would save disk space and be faster, allowing people to have gguf files in multiple places without paying the disk space penalty. This is only possible on a few different file systems. APFS is one (on all modern Macs), Btrfs, zfs, etc. Changing `cp` to `cp -c` should do it on Macs. `cp --reflink=auto` should work on linux.
GiteaMirror added the feature request label 2026-04-28 11:40:36 -05:00
GiteaMirror commented 2026-04-28 11:40:40 -05:00
Author
Owner
Copy Link

@spott commented on GitHub (May 24, 2024):

I would be happy to try and implement this, however I'm not sure where to find this part of the code base. If someone could point me in the right direction, I would appreciate it.

<!-- gh-comment-id:2128375356 --> @spott commented on GitHub (May 24, 2024): I would be happy to try and implement this, however I'm not sure where to find this part of the code base. If someone could point me in the right direction, I would appreciate it.
GiteaMirror commented 2026-04-28 11:40:42 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (May 27, 2024):

Ollama is client-server over HTTP. ollama create sends data over http to server. Server might reside on another machine from client, run as a different user with different permissions, or model storage might be on another filesystem. So, file system copy isn't used and wouldn't be general enough to cover all the use cases.

<!-- gh-comment-id:2133891817 --> @easp commented on GitHub (May 27, 2024): Ollama is client-server over HTTP. `ollama create` sends data over http to server. Server might reside on another machine from client, run as a different user with different permissions, or model storage might be on another filesystem. So, file system copy isn't used and wouldn't be general enough to cover all the use cases.
GiteaMirror commented 2026-04-28 11:40:46 -05:00
Author
Owner
Copy Link

@spott commented on GitHub (May 29, 2024):

I imagine the vast majority of users using ollama use it locally (though maybe I'm wrong about this, regardless I imagine it is a lot of users), so accepting a file path if it is known that the server is local would be an option.

Another option is to create a special "server side create" script that could be run.

So, file system copy isn't used and wouldn't be general enough to cover all the use cases.

Yes, this would be a special case optimization to drastically speed up creating new models and to save users gigabytes of disk space. It would only work on specific filesystems anyways. It isn't going to change the world, but it would be a welcome change that would help people.

<!-- gh-comment-id:2138231695 --> @spott commented on GitHub (May 29, 2024): I imagine the vast majority of users using ollama use it locally (though maybe I'm wrong about this, regardless I imagine it is a lot of users), so accepting a file path if it is known that the server is local would be an option. Another option is to create a special "server side create" script that could be run. > So, file system copy isn't used and wouldn't be general enough to cover all the use cases. Yes, this would be a special case optimization to drastically speed up creating new models and to save users gigabytes of disk space. It would only work on specific filesystems anyways. It isn't going to change the world, but it would be a welcome change that would help people.
GiteaMirror commented 2026-04-28 11:40:51 -05:00
Author
Owner
Copy Link

@lee-b commented on GitHub (Apr 21, 2025):

Personally, I have set up:

/models -- a GGUF directory on my server
/models -- a mount of that in my docker container

It would be really MUCH better if ollama simply copied the modelfile and hashed the existing image in this kind of setup, not copying/pulling the FROM file in cases where that FROM file already exists on the server.

With most other inference services/engines, this works fine, with zero-copy: I mount the folder, and tell the service/engine to run the file. When large model files are involved, copying becomes an issue in terms of time and disk space. Especially when experimenting with modelfile config etc.

<!-- gh-comment-id:2818043405 --> @lee-b commented on GitHub (Apr 21, 2025): Personally, I have set up: /models -- a GGUF directory on my server /models -- a mount of that in my docker container It would be really MUCH better if ollama simply copied the modelfile and hashed the existing image in this kind of setup, not copying/pulling the FROM file in cases where that FROM file already exists on the server. With most other inference services/engines, this works fine, with zero-copy: I mount the folder, and tell the service/engine to run the file. When large model files are involved, copying becomes an issue in terms of time and disk space. Especially when experimenting with modelfile config etc.
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 feature request
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#49399
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?