Small bugs/unexpected behavior and feedback with suggestions about llmfarm HOT 6 CLOSED

Hi. Thanks for your feedback.
I know about the keyboard closing bug, but it was not a quick fix, so I've postponed solving it for later.
I didn't know about the bug with deleting chats. Thank you.
I will try to add the setting to disable scrolling in the new version.
The memory usage is a good idea, I've been planning to add warnings to users that the model may not work on their device for a long time, but I like your idea better.
The button to clear chat will be moved to the toolbox in the new version.
Regarding customization templates, I'm thinking of creating a resource where users can share their templates. Add more templates available out of the box, I think it is pointless.
About your suggestions. Thank you very much for giving me ideas. Hopefully someday I will realize them all in some form.

from llmfarm.

Thank you for your answer. Yeah, I don’t realize there is custom prompt, so I think it’s enough. But I find something strange behavior. I use this model:

https://huggingface.co/TheBloke/OpenHermes-2-Mistral-7B-GGUF

And the prompt template is:
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

So i use:

<|im_start|>system
{Please meticulously review the provided text for any grammatical errors or inaccuracies, correcting them with your expertise as a professional English language corrector. Additionally, if you believe there are opportunities to enhance the text for better clarity or expression, feel free to make thoughtful adjustments to improve its overall quality.}<|im_end|>
<|im_start|>You
{prompt}<|im_end|>
<|im_start|>LLM

But it doesn’t work. After I try use double {{text}} and it works. But on the hugging face there is the prompt template only with one {text}. Do your app change something internally?

Is there any setting to limit the models answer length? I think I didn’t find any.

And one small suggestion. It will be nice to have settings to don’t turn off display while the app generate text.

Have a nice weekend!

from llmfarm.

And one small suggestion. I tried use model for coding. It use often triple backticks (```) before and after the text. But in your app it doesn’t put the text on block without formatting.

And small bug, copy text from the chat window doesn’t copy anything.

from llmfarm.

I only want to report, copy generated text still doesn't work in the latest beta. And the text scrolling and you just wait until end if you want to read the beginning. It will be nice, if we can control the maximum text model can generate, like now it's nice it's super long, but maybe sometimes I want to limit it. Because now, when the answer is long, you need to wait longer until you can read it from beginning.

from llmfarm.

Could you please clarify what you mean by copying? The function itself should be active, selection is not working yet.

from llmfarm.

It's active, but doesn't work. https://github.com/guinmoon/LLMFarm/assets/59890223/389e8603-41d8-4a49-8192-836264e11973
And the keyboard still cannot be hidden.

from llmfarm.