app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Initialize Hugging Face Inference Client
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# Response Function (Now Compatible with 'type=messages')
def respond(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Correct message format for Gradio's 'messages' type
    messages = [{"role": "system", "content": system_message}]

    # Handle both old tuple format and new 'messages' format
    for entry in history:
        if isinstance(entry, dict) and "role" in entry and "content" in entry:
            messages.append(entry)  # Already in correct format
        elif isinstance(entry, tuple) and len(entry) == 2:
            messages.append({"role": "user", "content": entry[0]})
            messages.append({"role": "assistant", "content": entry[1]})

    # Add the current user message
    messages.append({"role": "user", "content": message})

    # Initialize response string
    response = ""

    # Generate chat response using the client
    for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content
        response += token
        yield response

# Gradio Interface Setup
demo = gr.ChatInterface(
    respond,
    chatbot=gr.Chatbot(type="messages"),
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
    ],
)

# Launch the Gradio app
if __name__ == "__main__":
    demo.launch()

# Fine-Tuning GPT-2 on Hugging Face Spaces (Streaming 40GB Dataset, No Storage Issues)
from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
from datasets import load_dataset
from peft import LoraConfig, get_peft_model
import torch

# Authenticate Hugging Face
from huggingface_hub import notebook_login
notebook_login()

# Load GPT-2 model and tokenizer
model_name = "gpt2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Custom Dataset (Predefined Q&A Pairs for Project Expo)
custom_data = [
    {"prompt": "Who are you?", "response": "I am Eva, a virtual voice assistant."},
    {"prompt": "What is your name?", "response": "I am Eva, how can I help you?"},
    {"prompt": "What can you do?", "response": "I can assist with answering questions, searching the web, and much more!"},
    {"prompt": "Who invented the computer?", "response": "Charles Babbage is known as the father of the computer."},
    {"prompt": "Tell me a joke.", "response": "Why don’t scientists trust atoms? Because they make up everything!"},
    {"prompt": "Who is the Prime Minister of India?", "response": "The current Prime Minister of India is Narendra Modi."},
    {"prompt": "Who created you?", "response": "I was created by an expert team specializing in AI fine-tuning and web development."}
]

# Convert custom dataset to Hugging Face Dataset
dataset_custom = load_dataset("json", data_files={"train": custom_data})

# Load OpenWebText dataset (5% portion to avoid streaming issues)
dataset = load_dataset("Skylion007/openwebtext", split="train[:20%]")

# Tokenization function
def tokenize_function(examples):
    return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=512)

tokenized_datasets = dataset.map(tokenize_function, batched=True)

# Apply LoRA for efficient fine-tuning
lora_config = LoraConfig(
    r=8, lora_alpha=32, lora_dropout=0.05, bias="none",
    target_modules=["c_attn", "c_proj"]  # Apply LoRA to attention layers
)

model = get_peft_model(model, lora_config)
model.gradient_checkpointing_enable()  # Enable checkpointing for memory efficiency

# Training arguments
training_args = TrainingArguments(
    output_dir="gpt2_finetuned",
    auto_find_batch_size=True,
    gradient_accumulation_steps=4,
    learning_rate=5e-5,
    num_train_epochs=3,
    save_strategy="epoch",
    logging_dir="logs",
    bf16=True,
    push_to_hub=True
)

# Trainer setup
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets
)

# Start fine-tuning
trainer.train()

# Save and push the model to Hugging Face Hub
trainer.save_model("gpt2_finetuned")
tokenizer.save_pretrained("gpt2_finetuned")
trainer.push_to_hub()

# Deploy as Gradio Interface
def generate_response(prompt):
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_length=100)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

demo = gr.Interface(fn=generate_response, inputs="text", outputs="text")
demo.launch()