app.py

import gradio as gr
from huggingface_hub import InferenceClient

"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")


def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    response = ""

    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


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    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)",
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()

# Fine-Tuning GPT-2 on Hugging Face Spaces (Streaming 40GB Dataset, No Storage Issues)

# Install required libraries
!pip install transformers datasets peft accelerate bitsandbytes torch torchvision torchaudio gradio -q

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)

# Load the OpenWebText dataset using streaming (No download required)
dataset = load_dataset("Skylion007/openwebtext", split="train[:5%]")  # Load 5% to avoid streaming issues

# 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)

# Enable gradient checkpointing to reduce memory usage
model.gradient_checkpointing_enable()

# 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()