rabidcopy

> I have suggested a change at [rabidcopy#2](https://github.com/rabidcopy/llama.cpp/pull/2) (please check it out @rabidcopy) that would return control to the user by injecting the anti-prompt instead, which should solve that problem....

Ohh, this merge conflicts with https://github.com/ggerganov/llama.cpp/pull/330. Or rather doesn't work as the antiprompt is no longer tokenized.

> > Ohh, this merge conflicts with #330. Or rather doesn't work as the antiprompt is no longer tokenized. > > Oops! What if we tokenize the (first) reverse prompt...

I think this works. ``` Assistant: Hi! I'm your personal assistant to answer any questions you may have! You: Hi, sorry but I gotta go actually, unexpected plans. Goodbye! Assistant:...

> @eiz It seems there is a problem with the alpaca 13B, after conversion, when loading it complains about the embedding size: > > ``` > main: seed = 1679320340...

I think I'm going to throw together a personal hack that throws characters like `#()\/[]{}` out of sampling by piggybacking off the `-ignore-eos` code. I've been getting a lot more...

Haha. This is a bit overkill but it works. ``` You:Hi Assistant: Do you need anything? You: Can you type a backslash? Assistant: Sure thing! You: A forward slash? Assistant:...

` -c N, --ctx_size N size of the prompt context (default: 512)` You'll want to set `-c 2048` (max recommended for LlaMa) Also set `-n 2048` as well, effectively they...

I think the conversion script should be included in the repository somewhere to be honest. It's worked for every old format model I've thrown at it and saved me a...

Vicuna is a pretty strict model in terms of following that ### Human/### Assistant format when compared to alpaca and gpt4all. Less flexible but fairly impressive in how it mimics...