app.py

import gradio as gr
from transformers import pipeline, AutoTokenizer
import torch
import spaces
import os

@spaces.GPU
def load_model(model_name):
    return pipeline("text-generation", model=model_name, device="cuda", torch_dtype=torch.float16, trust_remote_code=True, token=os.environ["token"])
@spaces.GPU
def generate(
    model_name,
    user_input,
    temperature=0.4,
    top_p=0.95,
    top_k=50,
    max_new_tokens=256,
):
    pipe = load_model(model_name)

    # Set tokenize correctly. Otherwise ticking the box breaks it.
    prompt = f"<|im_start|>user\n{user_input}<|im_end|>\n<|im_start|>assistant\n"
    outputs = pipe(prompt, max_new_tokens=max_new_tokens, do_sample=True,
                   temperature=temperature, top_k=top_k, top_p=top_p, repetition_penalty=1.10)
    return outputs[0]["generated_text"]

model_choices = ["Locutusque/UltraQwen-7B", "Locutusque/UltraQwen-1_8B", "Locutusque/TinyMistral-248M-v2.5-Instruct", "M4-ai/TinyMistral-6x248M-Instruct", "Locutusque/Hercules-1.0-Mistral-7B", "Locutusque/Hercules-2.0-Mistral-7B"]
# What at the best options? 
g = gr.Interface(
    fn=generate,
    inputs=[
        gr.components.Dropdown(choices=model_choices, label="Model", value=model_choices[0], interactive=True),
        gr.components.Textbox(lines=2, label="Prompt", value="Write me a Python program that calculates the factorial of a given number."),
        gr.components.Slider(minimum=0, maximum=1, value=0.4, label="Temperature"),
        gr.components.Slider(minimum=0, maximum=1, value=0.95, label="Top p"),
        gr.components.Slider(minimum=0, maximum=100, step=1, value=50, label="Top k"),
        gr.components.Slider(minimum=1, maximum=1024, step=1, value=256, label="Max tokens"),  
    ],
    outputs=[gr.Textbox(lines=10, label="Output")],
    title="Locutusque's Language Models",
    description="Try out Locutusque's language models here! Credit goes to Mediocreatmybest for this space.",
    concurrency_limit=1
)

g.launch(max_threads=2)