github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9750] prefer offloading model layers over kv cache when both do not fit in VRAM. #68427

New Issue
Open
opened 2026-05-04 13:55:39 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-05-04 13:55:39 -05:00
Owner
Copy Link

Originally created by @apt-install-coffee on GitHub (Mar 14, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9750

Currently when running a model and large context size which do not both fit entirely in VRAM, ollama will reduce the number of model layers significantly instead of keeping the kv cache in main memory; for example by using --no-kv-offload.

My experiments when running a QwQ-32B quantized to IQ3_M (~14GB) at num_ctx=131072, on a 16GB 7800XT show significant improvements to token generation (reduced time thinking from ~2 hours to 35 minutes), when applying --no-kv-offload and allowing all model layers to be offloaded.

Originally created by @apt-install-coffee on GitHub (Mar 14, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9750 Currently when running a model and large context size which do not both fit entirely in VRAM, ollama will reduce the number of model layers significantly instead of keeping the kv cache in main memory; for example by using `--no-kv-offload`. My experiments when running a QwQ-32B quantized to IQ3_M (~14GB) at num_ctx=131072, on a 16GB 7800XT show significant improvements to token generation (reduced time thinking from ~2 hours to 35 minutes), when applying `--no-kv-offload` and allowing all model layers to be offloaded.
GiteaMirror added the feature request label 2026-05-04 13:55:39 -05:00
GiteaMirror commented 2026-05-04 13:55:41 -05:00
Author
Owner
Copy Link

@apt-install-coffee commented on GitHub (Mar 14, 2025):

I've opened #9751 to discuss what tweaking this manually might look like.

<!-- gh-comment-id:2723303101 --> @apt-install-coffee commented on GitHub (Mar 14, 2025): I've opened #9751 to discuss what tweaking this manually might look like.
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#68427
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?