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_map="auto", torch_dtype=torch.bfloat16, 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.
    if model_name == "M4-ai/tau-1.8B":
        prompt = user_input
    else:
        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)
    return outputs[0]["generated_text"]

model_choices = ["Locutusque/OpenCerebrum-1.0-7B-beta", "M4-ai/NeuralReyna-Mini-1.8B-v0.2", "Locutusque/Hyperion-3.0-Mistral-7B-DPO", "Locutusque/Hyperion-3.0-Mistral-7B-alpha", "M4-ai/tau-1.8B", "Locutusque/Hercules-4.0-Mistral-v0.2-7B", "Locutusque/Hercules-2.5-Mistral-7B", "M4-ai/tau-0.5B"]
# 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=2048, step=1, value=1024, label="Max tokens"),  
    ],
    outputs=[gr.Textbox(lines=10, label="Output")],
    title="Locutusque's Language Models",
    description="Try out Locutusque's (or other's) language models here! Credit goes to Mediocreatmybest for this space. You may also find some experimental preview models that have not been made public here.",
    concurrency_limit=1
)

g.launch(max_threads=4)