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 #2713] llava13b memory access faults on api/chat (firts call fine, fail on second one) #1625

New Issue
Closed
opened 2026-04-12 11:33:28 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-12 11:33:28 -05:00
Owner
Copy Link

Originally created by @uneuro on GitHub (Feb 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2713

image
I have 2x7900xtx

if I close ollama after each requests and specify only 1 gpu it's running well.
I tried 8 times to run ollama server and close after a request, at some point it was broken too cause closing wasn't clearing the vram
image

Originally created by @uneuro on GitHub (Feb 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2713 ![image](https://github.com/ollama/ollama/assets/5337885/74e03e82-5748-41c0-ab13-e18e1b102e56) I have 2x7900xtx if I close ollama after each requests and specify only 1 gpu it's running well. I tried 8 times to run ollama server and close after a request, at some point it was broken too cause closing wasn't clearing the vram <img width="1878" alt="image" src="https://github.com/ollama/ollama/assets/5337885/f64e242b-14d5-4bb9-a741-f425db2cc4e4">
GiteaMirror added the bugneeds more info labels 2026-04-12 11:33:28 -05:00
GiteaMirror commented 2026-04-12 11:33:29 -05:00
Author
Owner
Copy Link

@uneuro commented on GitHub (Feb 25, 2024):

played with https://github.com/ollama/ollama/pull/2146 keepalive parameter to 0 and had no more success (some vram still not free after shutdow from the keepalive).
tried today with same os, gpu on my desk with only 1 gpu and working like a charm so I suggest search arround the 2 gpu memory management, also noticing the vram going in the 2 gpu but 24gb vram cards and only 4-8 gb models

<!-- gh-comment-id:1962971567 --> @uneuro commented on GitHub (Feb 25, 2024): played with https://github.com/ollama/ollama/pull/2146 keepalive parameter to 0 and had no more success (some vram still not free after shutdow from the keepalive). tried today with same os, gpu on my desk with only 1 gpu and working like a charm so I suggest search arround the 2 gpu memory management, also noticing the vram going in the 2 gpu but 24gb vram cards and only 4-8 gb models
GiteaMirror commented 2026-04-12 11:33:30 -05:00
Author
Owner
Copy Link

@uneuro commented on GitHub (Feb 25, 2024):

got it again but on mistral this time, so the issue is related to 2xgpu more than llava (happens after a couple of working attempts)
image

<!-- gh-comment-id:1963017445 --> @uneuro commented on GitHub (Feb 25, 2024): got it again but on mistral this time, so the issue is related to 2xgpu more than llava (happens after a couple of working attempts) <img width="1874" alt="image" src="https://github.com/ollama/ollama/assets/5337885/87134b0f-1f77-42e2-9efa-8826bd6cf166">
GiteaMirror commented 2026-04-12 11:33:30 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Nov 6, 2024):

There's a reasonable chance the underlying defect has already been resolved. Can you retry on the latest release and let us know if you still see the problem?

<!-- gh-comment-id:2460535613 --> @dhiltgen commented on GitHub (Nov 6, 2024): There's a reasonable chance the underlying defect has already been resolved. Can you retry on the latest release and let us know if you still see the problem?
GiteaMirror commented 2026-04-12 11:33:31 -05:00
Author
Owner
Copy Link

@lpuv commented on GitHub (Nov 27, 2024):

I have the same issue with the latest version, 6700XT via thunderbolt egpu.

<!-- gh-comment-id:2502425839 --> @lpuv commented on GitHub (Nov 27, 2024): I have the same issue with the latest version, 6700XT via thunderbolt egpu.
GiteaMirror commented 2026-04-12 11:33:32 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Dec 3, 2024):

@lpuv eGPU support is tracked via #5411

<!-- gh-comment-id:2515593368 --> @dhiltgen commented on GitHub (Dec 3, 2024): @lpuv eGPU support is tracked via #5411
GiteaMirror commented 2026-04-12 11:33:33 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Dec 19, 2024):

I'm going to go ahead and close this since it's pretty stale and may already be fixed. As @dhiltgen mentioned we can track the eGPU issue separately.

<!-- gh-comment-id:2555816095 --> @pdevine commented on GitHub (Dec 19, 2024): I'm going to go ahead and close this since it's pretty stale and may already be fixed. As @dhiltgen mentioned we can track the eGPU issue separately.
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 needs more info
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#1625
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?