[GH-ISSUE #4816] deepseek v2 lite and 236b gets crazy after some conversations #3042

Open
opened 2026-04-12 13:28:02 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @netspym on GitHub (Jun 4, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4816

What is the issue?

Hi Ollama team:

I loaded the modes and keep them stay in memory. After some conversations, the models start to response with strange text and codes. After restarting ollama in ubuntu, the issue is gone.

OS

Linux

GPU

No response

CPU

AMD

Ollama version

0.1.40

Originally created by @netspym on GitHub (Jun 4, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4816 ### What is the issue? Hi Ollama team: I loaded the modes and keep them stay in memory. After some conversations, the models start to response with strange text and codes. After restarting ollama in ubuntu, the issue is gone. ### OS Linux ### GPU _No response_ ### CPU AMD ### Ollama version 0.1.40
GiteaMirror added the bug label 2026-04-12 13:28:02 -05:00
Author
Owner

@crassicus commented on GitHub (Jun 5, 2024):

I confirm the model not working in Linux ( at least in arch, binarie downloaded from core )

<!-- gh-comment-id:2148692313 --> @crassicus commented on GitHub (Jun 5, 2024): I confirm the model not working in Linux ( at least in arch, binarie downloaded from core )
Author
Owner

@netspym commented on GitHub (Jun 5, 2024):

And the speed is very slow giving that it is a 21b active params model. The speeds looks like a 70b model.

<!-- gh-comment-id:2149069368 --> @netspym commented on GitHub (Jun 5, 2024): And the speed is very slow giving that it is a 21b active params model. The speeds looks like a 70b model.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#3042