[GH-ISSUE #2689] Gemma model quantization or implementation seems botched #27364

Closed
opened 2026-04-22 04:39:54 -05:00 by GiteaMirror · 4 comments
Owner

Originally created by @horiacristescu on GitHub (Feb 22, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2689

I tried the gemma model today and it responds with

  • inconsistent formatting, such as using two commas instead of one
  • inconsistent phrasing, such as a noun not being at plural when it should, or absurd phrases like "copyright infringement is being violated"

I tried the same model on labs.perplexity.ai and their version seems to be coherent.

I used Gemma:7b-Instruct-Q5_K_M

Originally created by @horiacristescu on GitHub (Feb 22, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2689 I tried the gemma model today and it responds with - inconsistent formatting, such as using two commas instead of one - inconsistent phrasing, such as a noun not being at plural when it should, or absurd phrases like "copyright infringement is being violated" I tried the same model on labs.perplexity.ai and their version seems to be coherent. I used `Gemma:7b-Instruct-Q5_K_M`
Author
Owner

@penguinpatroller commented on GitHub (Feb 22, 2024):

This is probably the same issue as #2650. Still unresolved.

<!-- gh-comment-id:1960296949 --> @penguinpatroller commented on GitHub (Feb 22, 2024): This is probably the same issue as #2650. Still unresolved.
Author
Owner

@jmorganca commented on GitHub (Feb 23, 2024):

Hi there, sorry about this. This should be fixed now. You can de-download the latest models by running:

ollama pull gemma
ollama pull gemma:7b # same as above
ollama pull gemma:2b
<!-- gh-comment-id:1960597023 --> @jmorganca commented on GitHub (Feb 23, 2024): Hi there, sorry about this. This should be fixed now. You can de-download the latest models by running: ``` ollama pull gemma ollama pull gemma:7b # same as above ollama pull gemma:2b ```
Author
Owner

@mathiscode commented on GitHub (Mar 3, 2024):

I was still experiencing this with the latest model of gemma:7b-instruct and completely killing ollama and restarting seems to have fixed the issue for me.

<!-- gh-comment-id:1975051568 --> @mathiscode commented on GitHub (Mar 3, 2024): I was still experiencing this with the latest model of gemma:7b-instruct and completely killing ollama and restarting seems to have fixed the issue for me.
Author
Owner

@Gautamshahi commented on GitHub (Aug 4, 2024):

How did you check the quantization? I couldn't find in model description https://ollama.com/library/gemma:7b-instruct-q3_K_M/blobs/bc1362d999cf

<!-- gh-comment-id:2267816215 --> @Gautamshahi commented on GitHub (Aug 4, 2024): How did you check the quantization? I couldn't find in model description https://ollama.com/library/gemma:7b-instruct-q3_K_M/blobs/bc1362d999cf
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#27364