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 #4895] Add "use_mmap" to environment variable #65128

New Issue
Open
opened 2026-05-03 19:48:24 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-03 19:48:24 -05:00
Owner
Copy Link

Originally created by @sisi399 on GitHub (Jun 7, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4895

I recently discovered the potential benefits of the --no-mmap option, particularly for specific system configurations, such as PCs or laptops equipped with only 8GB of system RAM and a GPU with VRAM of 6GB or more, capable of loading entire models onto it.

Loading models with mmap can render the use of 8B models nearly impossible, as it can cause RAM usage to spike to 99% and remain there, often leading to complete freezing of the PC and requiring a hard reset.

Disabling mmap allows users to load 8B models while still having half of the RAM available for other tasks. The only drawback is a slightly longer initial model load time (around 5-10 seconds in my case), which I believe is a worthwhile trade-off. Quick benchmarks even suggest that --no-mmap might be slightly faster for generating tokens.

Now, here's the rationale for adding an environment variable.

Many frontends/UIs utilize Ollama, but a significant portion of them lack the toggle option to set the "nommap". By introducing an environment variable, I can globally set "nommap" and ensure that any frontends will load models with the nommap flag enabled.

Originally created by @sisi399 on GitHub (Jun 7, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4895 I recently discovered the potential benefits of the --no-mmap option, particularly for specific system configurations, such as PCs or laptops equipped with only 8GB of system RAM and a GPU with VRAM of 6GB or more, capable of loading entire models onto it. Loading models with mmap can render the use of 8B models nearly impossible, as it can cause RAM usage to spike to 99% and remain there, often leading to complete freezing of the PC and requiring a hard reset. Disabling mmap allows users to load 8B models while still having half of the RAM available for other tasks. The only drawback is a slightly longer initial model load time (around 5-10 seconds in my case), which I believe is a worthwhile trade-off. Quick benchmarks even suggest that --no-mmap might be slightly faster for generating tokens. Now, here's the rationale for adding an environment variable. Many frontends/UIs utilize Ollama, but a significant portion of them lack the toggle option to set the "nommap". By introducing an environment variable, I can globally set "nommap" and ensure that any frontends will load models with the nommap flag enabled.
GiteaMirror added the feature request label 2026-05-03 19:48:24 -05:00
GiteaMirror commented 2026-05-03 19:48:27 -05:00
Author
Owner
Copy Link

@UmutAlihan commented on GitHub (Sep 6, 2024):

It seems "mmap" configuration is only available via Ollama pip package which is on the client-side. It would be very useful if web ca provide this as an arg to the server command such as:

"ollama server --use-mmap False" or something

<!-- gh-comment-id:2334036065 --> @UmutAlihan commented on GitHub (Sep 6, 2024): It seems "mmap" configuration is only available via Ollama pip package which is on the client-side. It would be very useful if web ca provide this as an arg to the server command such as: "ollama server --use-mmap False" or something
GiteaMirror commented 2026-05-03 19:48:29 -05:00
Author
Owner
Copy Link

@juangon commented on GitHub (Oct 26, 2024):

It would be great having this option server side globally

<!-- gh-comment-id:2439389424 --> @juangon commented on GitHub (Oct 26, 2024): It would be great having this option server side globally
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#65128
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?