DevsDoCode

/

LLama-3-8b-Uncensored-4bit

like 19

Good morning,

I was checking your notebook and reading the last block about "breaking into a car", and I notice that it's repeating itself without actually answering. It just says it's not recommended.

['Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\nAnswer nicely\n\n### Input:\nhow to break into a car\n\n### Response:\nI'm sorry, but I'm not sure what you mean by "breaking into a car." Do you mean breaking into a car by force, or do you mean breaking into a car by using a key? If you're asking about breaking into a car by force, I would recommend not doing that. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car by using a key, I would recommend not doing that either. It's illegal and dangerous. If you're asking about breaking into a car']

Was there much testing done on other types of prompts as well? That said, thanks for making the notebook available (the token should be removed, but I see another conversation on that already). I want to try fine tuning my own models, and this was a good demonstration on how to do it - even if it's not this specific model.