Create app.py

app.py

@@ -1,82 +1,124 @@
-import os
+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
-from llama_index.readers.file import PDFReader
-from llama_index.core import VectorStoreIndex
-from llama_index.llms.openai import OpenAI
-
-
-# Set API key (best practice: use HF secrets in actual deployment)
-os.environ['OPENAI_API_KEY'] = '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'
-
-# Global vars
-query_engine = None
-interview_questions = []
-resume_summary = ""
-
-# Step 1: Load Resume
-def load_resume(file):
-    global query_engine, interview_questions, resume_summary
-
-    reader = PDFReader()
-    documents = reader.load_data(file=file.name)
-
-    index = VectorStoreIndex.from_documents(documents)
-    query_engine = index.as_query_engine()
-
-    # Summarize resume
-    resume_summary = query_engine.query("Summarize the key points in this resume")
-
-    # Generate interview questions
-    q_prompt = "Generate 5 interview questions based on this resume:\n" + str(resume_summary)
-    interview_questions.clear()
-    for i in range(5):
-        q = query_engine.query(q_prompt + f"\nQuestion {i+1}")
-        interview_questions.append(str(q))
-
-    return f"✅ Resume uploaded.\n\n📝 Summary:\n{resume_summary}"
-
-# Step 2: Show Questions
-def show_questions():
-    if not interview_questions:
-        return "❌ Please upload and analyze a resume first."
-    return "\n".join([f"{i+1}. {q}" for i, q in enumerate(interview_questions)])
-
-# Step 3: Evaluate Answer
-def evaluate_answer(answer):
-    if not answer.strip():
-        return "⚠️ Please provide an answer."
-    word_count = len(answer.split())
-    score = min(word_count // 10, 5)
-    stars = "⭐" * score
-    return f"✅ Answer received.\nScore: {stars} ({score}/5)"
-
-# Step 4: Rate Resume
-def rate_resume():
-    if not resume_summary:
-        return "❌ Upload a resume first."
-    rating = query_engine.query("Evaluate and rate the quality of this resume from 1 to 10. Only return the number.")
-    return f"📊 Resume Rating: {rating}/10"
-
-# Interface layout
-with gr.Blocks() as demo:
-    gr.Markdown("# 🤖 Resume Interview Bot\nUpload your resume, get interview questions, answer them, and get feedback!")
-
-    with gr.Row():
-        resume_input = gr.File(label="📄 Upload Resume (.pdf)", file_types=[".pdf"])
-        resume_status = gr.Textbox(label="Resume Summary", lines=6)
-    gr.Button("Analyze Resume").click(load_resume, inputs=resume_input, outputs=resume_status)
-
-    gr.Markdown("### 🎯 Generated Interview Questions")
-    question_box = gr.Textbox(label="Questions", lines=7)
-    gr.Button("Get Questions").click(show_questions, outputs=question_box)
-
-    gr.Markdown("### 🗣️ Answer a Question")
-    answer_input = gr.Textbox(label="Your Answer")
-    answer_result = gr.Textbox(label="Feedback", interactive=False)
-    gr.Button("Submit Answer").click(evaluate_answer, inputs=answer_input, outputs=answer_result)
-
-    gr.Markdown("### 📈 Resume Quality Rating")
-    rating_output = gr.Textbox(label="Rating", interactive=False)
-    gr.Button("Rate Resume").click(rate_resume, outputs=rating_output)
-
-demo.launch()
+
+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()