[GH-ISSUE #4062] llama3-gradient going crazy #49033

Closed
opened 2026-04-28 10:37:19 -05:00 by GiteaMirror · 8 comments
Owner

Originally created by @DuckyBlender on GitHub (Apr 30, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4062

What is the issue?

Just ran ollama run llama3-gradient and this happened. Used the default (2k) context size
image

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.1.32

Originally created by @DuckyBlender on GitHub (Apr 30, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4062 ### What is the issue? Just ran `ollama run llama3-gradient` and this happened. Used the default (2k) context size ![image](https://github.com/ollama/ollama/assets/42645784/6c390484-774a-40c4-82fe-4411d65b471c) ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.32
GiteaMirror added the bug label 2026-04-28 10:37:19 -05:00
Author
Owner

@jmorganca commented on GitHub (Apr 30, 2024):

Hi there, the runtime parameters have been updated to help with this issue when the context limit is hit. Please try re-pulling using ollama pull llama3-gradient which should fix this.

<!-- gh-comment-id:2087605644 --> @jmorganca commented on GitHub (Apr 30, 2024): Hi there, the runtime parameters have been updated to help with this issue when the context limit is hit. Please try re-pulling using `ollama pull llama3-gradient` which should fix this.
Author
Owner

@DuckyBlender commented on GitHub (Apr 30, 2024):

After re-pulling the model it seems to be working :)

<!-- gh-comment-id:2087678216 --> @DuckyBlender commented on GitHub (Apr 30, 2024): After re-pulling the model it seems to be working :)
Author
Owner

@jmorganca commented on GitHub (May 1, 2024):

@DuckyBlender great to hear, and sorry again about that.

<!-- gh-comment-id:2087884422 --> @jmorganca commented on GitHub (May 1, 2024): @DuckyBlender great to hear, and sorry again about that.
Author
Owner

@DuckyBlender commented on GitHub (May 2, 2024):

image
Everything seemed to be working perfectly, until it started repeating itself

<!-- gh-comment-id:2091307821 --> @DuckyBlender commented on GitHub (May 2, 2024): ![image](https://github.com/ollama/ollama/assets/42645784/2d52d264-e88e-4bd8-bbc4-610ca7361e64) Everything seemed to be working perfectly, until it started repeating itself
Author
Owner

@lazyracket commented on GitHub (May 4, 2024):

Is it related to the llama.cpp BPE? https://github.com/ollama/ollama/issues/4126

<!-- gh-comment-id:2094268660 --> @lazyracket commented on GitHub (May 4, 2024): Is it related to the `llama.cpp` BPE? https://github.com/ollama/ollama/issues/4126
Author
Owner

@binaryc0de commented on GitHub (Jun 18, 2024):

I'm seeing the same thing. llama3-gradient was working fine for a while and now it endlessly repeats it's self.

<!-- gh-comment-id:2177267729 --> @binaryc0de commented on GitHub (Jun 18, 2024): I'm seeing the same thing. llama3-gradient was working fine for a while and now it endlessly repeats it's self.
Author
Owner

@DuckyBlender commented on GitHub (Jun 26, 2024):

Maybe a higher repeat penalty can fix this?

<!-- gh-comment-id:2192322052 --> @DuckyBlender commented on GitHub (Jun 26, 2024): Maybe a higher repeat penalty can fix this?
Author
Owner

@step21 commented on GitHub (Nov 8, 2024):

image Everything seemed to be working perfectly, until it started repeating itself

I have the same thing. Worst thing is, that it doesn't stop but keeps going for a really long time, using resources.

<!-- gh-comment-id:2464799204 --> @step21 commented on GitHub (Nov 8, 2024): > ![image](https://private-user-images.githubusercontent.com/42645784/327557376-2d52d264-e88e-4bd8-bbc4-610ca7361e64.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.ETLVvceFKxlB6tX25h39i6ioUTd9juRjGgGLx0v3c3A) Everything seemed to be working perfectly, until it started repeating itself I have the same thing. Worst thing is, that it doesn't stop but keeps going for a really long time, using resources.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#49033