app.py

import gradio as gr
from huggingface_hub import Repository, InferenceClient
import os
import json
import re

API_TOKEN = os.environ.get("API_TOKEN")

API_ENDPOINTS = {
    "Falcon": "tiiuae/falcon-180B-chat",
    "Llama": "meta-llama/Llama-2-70b-chat-hf",
    "Mistral": "mistralai/Mistral-7B-v0.1",
    "Mistral*": "mistralai/Mistral-7B-Instruct-v0.1",
    "Open-3.5": "openchat/openchat_3.5",
    "Xistral": "mistralai/Mixtral-8x7B-v0.1",
    "Xistral*": "mistralai/Mixtral-8x7B-Instruct-v0.1",
}

CHOICES = []
CLIENTS = {}

for model_name, model_endpoint in API_ENDPOINTS.items():
    CHOICES.append(model_name)
    CLIENTS[model_name] = InferenceClient(model_endpoint, headers = { "Authorization": f"Bearer {API_TOKEN}" })

def predict(input, model, temperature, top_p, top_k, rep_p, max_tokens, stop_seqs, seed):
    stops = json.loads(stop_seqs)
   
    response = CLIENTS[model].text_generation(
        input,
        temperature = temperature,
        max_new_tokens = max_tokens,
        top_p = top_p,
        top_k = top_k,
        repetition_penalty = rep_p,
        stop_sequences = stops,
        do_sample = True,
        seed = seed,
        stream = False,
        details = False,
        return_full_text = False
    )

    return response

with gr.Blocks() as demo:
    with gr.Row(variant = "panel"):
        gr.Markdown("✡️ This is a private LLM Space owned within STC Holdings!")
            
    with gr.Row():
        with gr.Column():
            input = gr.Textbox(label = "Input", value = "", lines = 2)
            run = gr.Button("▶")
            
        with gr.Column():
            model = gr.Dropdown(choices = CHOICES, value = next(iter(API_ENDPOINTS)), interactive = True, label = "Model")
            temperature = gr.Slider( minimum = 0, maximum = 2, value = 1, step = 0.01, interactive = True, label = "Temperature" )
            top_p = gr.Slider( minimum = 0.01, maximum = 0.99, value = 0.95, step = 0.01, interactive = True, label = "Top P" )
            top_k = gr.Slider( minimum = 1, maximum = 2048, value = 50, step = 1, interactive = True, label = "Top K" )
            rep_p = gr.Slider( minimum = 0.01, maximum = 2, value = 1.2, step = 0.01, interactive = True, label = "Repetition Penalty" )
            max_tokens = gr.Slider( minimum = 1, maximum = 2048, value = 32, step = 64, interactive = True, label = "Max New Tokens" )
            stop_seqs = gr.Textbox( value = "", interactive = True, label = "Stop Sequences ( JSON Array / 4 Max )" )
            seed = gr.Slider( minimum = 0, maximum = 9007199254740991, value = 42, step = 1, interactive = True, label = "Seed" )
            
    with gr.Row():
        with gr.Column():
            output = gr.Textbox(label = "Output", value = "", lines = 50)

    run.click(predict, inputs = [input, model, temperature, top_p, top_k, rep_p, max_tokens, stop_seqs, seed], outputs = [output], queue = False)
    
demo.launch(show_api = True)