app.py

import torch
from datasets import Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer, BitsAndBytesConfig
from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training

model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=bnb_config,
    device_map="auto"
)

tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
tokenizer.pad_token = tokenizer.eos_token
model.config.use_cache = False
model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)

lora_config = LoraConfig(
    r=8,
    lora_alpha=32,
    target_modules=["q_proj", "v_proj"],
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM"
)

model = get_peft_model(model, lora_config)
import pandas as pd
from datasets import Dataset


# Load data from CSV
df = pd.read_csv("Customer-Support.csv")


# Rename columns to match expected keys
df = df.rename(columns={"query": "instruction", "response": "output"})


# Select required columns
data = df[["instruction", "output"]].fillna("")


# Convert DataFrame to list of dictionaries
data = data.to_dict(orient="records")


# Create Hugging Face Dataset
dataset = Dataset.from_list(data)


# Format each example
def format_instruction(example):
    return f"### Instruction:\n{example['instruction']}\n\n### Response:\n{example['output']}"


# Map formatted text
dataset = dataset.map(lambda x: {"text": format_instruction(x)})

def tokenize_function(example):
    tokenized = tokenizer(example["text"], truncation=True, padding="max_length", max_length=512)
    tokenized["labels"] = tokenized["input_ids"].copy()
    return tokenized

tokenized_dataset = dataset.map(tokenize_function, batched=True)
training_args = TrainingArguments(
    output_dir="./tinyllama-qlora-support-bot",
    per_device_train_batch_size=2,
    gradient_accumulation_steps=4,
    learning_rate=2e-4,
    logging_dir="./logs",
    num_train_epochs=3,
    logging_steps=10,
    save_total_limit=2,
    save_strategy="epoch",
    bf16=True,
    optim="paged_adamw_8bit"
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_dataset,
    tokenizer=tokenizer
)

trainer.train()
model.save_pretrained("tinyllama-qlora-support-bot")
tokenizer.save_pretrained("tinyllama-qlora-support-bot")

from transformers import pipeline

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

instruction = "How do I update the app?"
prompt = f"### Instruction:\n{instruction}\n\n### Response:\n"

output = pipe(prompt, max_new_tokens=100)
print(output[0]['generated_text'])
import gradio as gr

def generate_response(instruction):
    prompt = f"### Instruction:\n{instruction}\n\n### Response:\n"
    output = pipe(prompt, max_new_tokens=100, do_sample=True, temperature=0.7)
    # Extract only the response part
    response = output[0]["generated_text"].split("### Response:\n")[-1].strip()
    return response

gr.Interface(
    fn=generate_response,
    inputs=gr.Textbox(lines=3, label="Enter your question"),
    outputs=gr.Textbox(lines=5, label="Support Bot's Response"),
    title="📞 Customer Support Chatbot",
    description="Ask a question and get a response from your fine-tuned TinyLLaMA model.",
).launch()