Skip to main content

Does anyone else experience the same issue I'm having with Google's Gemma-7B on Ollama? Sometimes, it even struggles with spelling simple words in English and creates nonsensical words! This is not what I expected. #LLM #ML #AI #Ollama
#AI #ML #llm #ollama
in reply to Chi Kim

victor tsaran
victor tsaran
• •
Does it actually work for you? I am getting HTTP header errors whenever I try hitting it via curl, script or from CLI.
in reply to victor tsaran

victor tsaran
victor tsaran
• •
O wait, for some reason, my Ollama update arrived only now. Updated and am able to communicate with the model. I am also seeing a lot of gibberish in responses.
in reply to victor tsaran

Chi Kim
Chi Kim
• •
@vick21 I get ok response, but some spelling errors and made up words. Hopefully it's just a bug with quant or llama.cpp implementation.
@victor tsaran
in reply to Chi Kim

victor tsaran
victor tsaran
• •
I am not sure if I will keep it though. It’s too censored for my taste!
in reply to victor tsaran

Chi Kim
Chi Kim
• •
@vick21 Yeah. I deleted for now. Especially you can use Gemini pro for free anyways.
@victor tsaran
in reply to Chi Kim

victor tsaran
victor tsaran
• •
Also, I noticed that the 2b version does not have these artifacts.
in reply to victor tsaran

Chi Kim
Chi Kim
• •
@vick21 Oh interesting. It must be problem with llama.cpp or quant then.
@victor tsaran
⇧