Hugging Face universal adapter

[asigalov61]

@WebChatAppAi Hey bro,

Thank you for creating this project. Its very nice and has a lot of potential :)

I was wondering if you could implement a universal Hugging Face adapter for MIDI models.

There are several models already available and all they need is a nice frontend like yours.

As you know HF has a nice and simple requested API. I.e:

import { Client } from "@gradio/client";

const response_0 = await fetch("https://github.com/gradio-app/gradio/raw/main/test/test_files/sample_file.pdf");
const exampleFile = await response_0.blob();
						
const client = await Client.connect("asigalov61/Godzilla-Piano-Transformer");
const result = await client.predict("/generate_music_and_state", { 
				input_midi: exampleFile, 		
		num_prime_tokens: 15, 		
		num_gen_tokens: 15, 		
		num_mem_tokens: 15, 		
		model_temperature: 0.1, 
});

console.log(result.data);

It would be cool if your project featured a universal adapter with all of these models so that the users can choose which model to use. Since HF requires auth, there has to be a field for a AUTH TOKEN but otherwise, HF provides everything for free and even without auth.

I would be very interested in collaborating on such sub-project for your project. I am not good with frontends but I can do some stuff anyway.

Here is an example of what I did with one of my HF spaces (this is not AI but still).

https://asigalov61.github.io/

Let me know if you have time and interest in something like that.

Sincerely,

Alex.

[WebChatAppAi]

Hey Alex,

Awesome — you know what, I was actually exploring your Tegridy-Tools and it's awesome. I was about to reach out to you to discuss how we could implement a model into it.

And I have a great idea — I can give you GPU access if you need, like an NVIDIA L40s with 48GB VRAM. I want you to train a model for melody generation only, and some more stuff.

Let me know when we can discuss.

— Alvan

[asigalov61]

@WebChatAppAi I emailed you. Thank you :)

[WebChatAppAi]

Hey Alex,

I’ve added your Godzilla-Piano-Transformer as a plugin — pull the latest update and check it out.
It’s a bit slow right now because your HF Space is generating responses slowly — that’s actually why I hadn’t tried any of the API endpoints until now.
Also, since your API supports multiple batch types, the plugin currently allows selecting only one specific batch at a time through the config. You can test it with existing notes or an empty input — both should work fine.
Let me know what you think or if you'd like to expand support later.

— Alvan

[asigalov61]

@WebChatAppAi Thanks a lot. I will try it right now :)

The generation is actually pretty fast, it's the audio rendering that takes a while since its 12 batches of audio. But this can be fixed or adjusted as needed.

[WebChatAppAi]

let me know how is that

[WebChatAppAi]

@asigalov61 Hey Alex, hope you're doing well. Please pull the latest update and test the new model. Loading and generation, as you described, I have implemented auto-detect for model parameters. Let me know

Also, if you have any new ideas or designs for the AI Studio panel, let me know. Automatic model download or anything else—let me know, I will implement it as needed.

[asigalov61]

@WebChatAppAi I responded via email. Thank you.

[WebChatAppAi]

It will be added soon @asigalov61 i need u to give me specific url for model discovery

[WebChatAppAi]

pull latest updates and take a look the new download manager

[asigalov61]

@WebChatAppAi Thank you very much. Awesome work! :)

I am a bit busy now, but I will try it when I have some free time :)

All models for MuseCraft are located here: https://huggingface.co/projectlosangeles/MuseCraft

Alex.