app.py

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

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

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

SPECIAL_SYMBOLS_AI = ["ㅤ", "ㅤ"]
SPECIAL_SYMBOLS_USER = ["⠀", "⠀"] # ["‹", "›"] ['"', '"']

DEFAULT_INPUT = "User: Hi!"
DEFAULT_WRAP = "Statical: %s"
DEFAULT_INSTRUCTION = "Conversation: Statical is a helpful chatbot who is communicating with people."

DEFAULT_STOPS = '["ㅤ", "⠀"]' # '["‹", "›"]' '[\"\\\"\"]'

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",
    "Hermes": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
}

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 format(instruction, history, input, wrap):
    sy_la, sy_ra = SPECIAL_SYMBOLS_AI[0], SPECIAL_SYMBOLS_AI[1]
    sy_l, sy_r = SPECIAL_SYMBOLS_USER[0], SPECIAL_SYMBOLS_USER[1]
    wrapped_input = wrap % ("")
    formatted_history = "".join(f"{sy_l}{message[0]}{sy_r}{sy_la}{message[1]}{sy_la}" for message in history)
    formatted_input = f"{sy_la}{instruction}{sy_ra}{formatted_history}{sy_l}{input}{sy_r}{sy_la}"
    return f"{formatted_input}{wrapped_input}", formatted_input
    
def predict(access_key, instruction, history, input, wrap, model, temperature, top_p, top_k, rep_p, max_tokens, stop_seqs, seed):

    if (access_key != KEY):
        print(">>> MODEL FAILED: Input: " + input + ", Attempted Key: " + access_key)
        return ("[UNAUTHORIZED ACCESS]", input, []);

    instruction = instruction or DEFAULT_INSTRUCTION
    history = history or []
    input = input or ""
    wrap = wrap or ""
    stop_seqs = stop_seqs or DEFAULT_STOPS
        
    stops = json.loads(stop_seqs)

    formatted_input, formatted_input_base = format(instruction, history, input, wrap)
    
    print(seed)
    print(formatted_input)
    print(model)

    response = CLIENTS[model].text_generation(
        formatted_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
    )

    result = wrap % (response)
    
    for stop in stops:
        result = result.split(stop, 1)[0]
    for symbol in stops:
        result = result.replace(symbol, '')

    history = history + [[input, result]]

    print(f"---\nUSER: {input}\nBOT: {result}\n---")

    return (result, input, history)

def clear_history():
    print(">>> HISTORY CLEARED!")
    return []
     
def maintain_cloud():
    print(">>> SPACE MAINTAINED!")
    return ("SUCCESS!", "SUCCESS!")

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():
            history = gr.Chatbot(label = "History", elem_id = "chatbot")
            input = gr.Textbox(label = "Input", value = DEFAULT_INPUT, lines = 2)
            wrap = gr.Textbox(label = "Wrap", value = DEFAULT_WRAP, lines = 1)
            instruction = gr.Textbox(label = "Instruction", value = DEFAULT_INSTRUCTION, lines = 4)
            access_key = gr.Textbox(label = "Access Key", lines = 1)
            run = gr.Button("▶")
            clear = gr.Button("🗑️")
            cloud = 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 = DEFAULT_STOPS, 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 = [access_key, instruction, history, input, wrap, model, temperature, top_p, top_k, rep_p, max_tokens, stop_seqs, seed], outputs = [output, input, history], queue = False)
    clear.click(clear_history, [], history, queue = False)
    cloud.click(maintain_cloud, inputs = [], outputs = [input, output], queue = False)
    
demo.launch(show_api = True)