github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10631] Proposal for Manual VRAM Allocation in Modelfile #6994

New Issue
Open
opened 2026-04-12 18:53:16 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-12 18:53:16 -05:00
Owner
Copy Link

Originally created by @sunhy0316 on GitHub (May 9, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10631

PARAMETER num_gpu xxx

Originally posted by @rick-github in #10359

Hi Team,

I'd like to suggest adding a feature to manually specify the VRAM allocation size for models directly in the Modelfile. Currently, when loading a second model, the first model is forcibly unloaded even if there's sufficient remaining VRAM. While num_gpu ensures a single model stays entirely in VRAM, it doesn't guarantee coexistence for multiple models.

By allowing explicit VRAM size declaration in the Modelfile (e.g., vram_size 24GB), Ollama could:

  1. Compare the declared size against actual free VRAM. If adequate, load the new model without unloading the first.
  2. Maximize GPU utilization by enabling parallel model loading when possible.
  3. Unload other model if the declared size exceeds available VRAM (clear error messaging).

This approach would address:

  • The current conservative estimation issue forcing unnecessary unloads.
  • User control over resource allocation for multi-model workflows.
  • Efficient VRAM usage without compromising stability.

Would this align with Ollama's memory management roadmap? Looking forward to your thoughts.

Originally created by @sunhy0316 on GitHub (May 9, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10631 > ``` > PARAMETER num_gpu xxx > ``` _Originally posted by @rick-github in [#10359](https://github.com/ollama/ollama/issues/10359#issuecomment-2864661617)_ --- Hi Team, I'd like to suggest adding a feature to manually specify the VRAM allocation size for models directly in the Modelfile. Currently, when loading a second model, the first model is forcibly unloaded even if there's sufficient remaining VRAM. While `num_gpu` ensures a single model stays entirely in VRAM, it doesn't guarantee coexistence for multiple models. By allowing explicit VRAM size declaration in the Modelfile (e.g., `vram_size 24GB`), Ollama could: 1. Compare the declared size against actual free VRAM. If adequate, load the new model *without* unloading the first. 2. Maximize GPU utilization by enabling parallel model loading when possible. 3. Unload other model if the declared size exceeds available VRAM (clear error messaging). This approach would address: - The current conservative estimation issue forcing unnecessary unloads. - User control over resource allocation for multi-model workflows. - Efficient VRAM usage without compromising stability. Would this align with Ollama's memory management roadmap? Looking forward to your thoughts. ---
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
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#6994
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?