addyosmani / chatty Goto Github PK

As a user, I want to control the chat configuration dynamically.

Depending on the machine specs, the sliding_window_size and attention_sink_size may be needed to avoid a 'max window length exceeded' error for example.

Reference: https://github.com/mlc-ai/web-llm/blob/main/src/config.ts#L66

Working on getting this in soon.

I'm willing to work on this task.

And if this task requires some discussion, please let me know.

Thanks

Right now, we have these two following themes:

• light
• dark

Let's add one more, called system sync (and let's make it default maybe?!!)

Eager to work on this, so feel free to assign this to me.

I think a little improvement can be here, on the screenshot, when the screen size gets narrower (less than 400px) the layout becomes inconsistent.

I think this can be a better way to do it.

If you think too I already made a fix locally, I will just create a PR so it can be merged after approval.

Why is there no qwen2 model?

While trying to use the project from a non-technical user perspective I came across this:

I think if the project is going to be used by anyone who wants to use it, maybe a better way to tell if something went wrong would be via error codes or simple error message like: "Failed to load/download the model, Check your internet connection"

There's also the URL format:

What do you think about a URL in this format "chatty.com/c/[id_of_the_conversation]" ? I think it's more intuitive this way since we kind of entering a sub-section of the app.
Suggestion: We can replace the "c" by a "d" to say directory or discussion or even "p" for prompt

Let's provide users some electron build as well, so that we can get it like a native desktop application, and can run it without opening browsers.

Most importantly it will open lots of opportunities, like lots of desktop integrations, tools, etc.

1. Consider adding a "Getting Started" or "Installation" section that provides instructions on how to set up and run the project locally. This would be helpful for users who want to try out the project on their own machines.
2. We can add a brief "Contributing" section to encourage others to contribute to your project. Provide guidelines on how to submit bug reports, feature requests, or pull requests.
3. If we have any future plans (e.g. nice to haves we recently discussed), we could include a "Roadmap" or "Upcoming Features" section to give users an idea of what to expect in future releases. We could frame it as features we'd be open to having contributed alternatively.

Just so we have basic GA in place

I think dockerizing the application will help a lot. As you can simply run and destroy the container, without keeping any unwanted files in your system.

If it is worth it, I am willing to contribute on this one.

Thanks

I noticed when the screen sizes gets to 768px the "Change name" button is touching the input that it depends on.

got inspired by #37

I think that having a rename functionality will help user easily identify their chats.

This will be good UX as people are accustomed to keeping their own chat names instead of the first message.

every time I send a query, I get this same response back: failed to load model: No URL found for model ID "Text-based model". Please check if the model ID is correct and included in the model_list configuration.. I have tried different models.

related snap:

Initial chat will need to download the model, but there is no progress indicator.
When using a large model, as a user I don't know the progress.

The way we are currently handling instantiating the web-llm-helper class is suboptimal.
It causes a large initial bundle size when the app is first rendernes as seen on the image below.

(Using next/bundle-analyzer):

For now, this is fine as we're focusing on the desktop version for the current release. However, ideally, we would want this to be handled only when needed and not at initial render to optimize the bundle size.