Delete app.py

app.py

@@ -1,100 +0,0 @@
-import torch
-import streamlit as st
-from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, Trainer, DataCollatorForSeq2Seq
-from datasets import load_dataset
-from peft import LoraConfig, get_peft_model
-import os
-
-# UI
-st.title("AI Tutor (Fine-tuned LLM)")
-st.write("This AI tutor is fine-tuned on Python-related questions.")
-
-# Load base model and tokenizer
-model_name = "microsoft/phi-2"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-
-# 🔥 Fix: Add padding token if missing
-if tokenizer.pad_token is None:
-    tokenizer.add_special_tokens({'pad_token': '[PAD]'})
-
-# Check if fine-tuned model exists
-model_path = "./models"
-if os.path.exists(model_path):
-    st.write("✅ Loading fine-tuned model...")
-    model = AutoModelForCausalLM.from_pretrained(model_path)
-else:
-    st.write("⚡ Fine-tuning the model (this will take time)...")
-
-    # Load model on CPU
-    model = AutoModelForCausalLM.from_pretrained(
-        model_name,
-        torch_dtype=torch.float32,  # Use float32 for CPU compatibility
-        device_map={"": "cpu"}  # Force CPU usage
-    )
-
-    # Resize model embeddings
-    model.resize_token_embeddings(len(tokenizer))
-
-    # Apply LoRA
-    lora_config = LoraConfig(
-        r=8,
-        lora_alpha=32,
-        target_modules=["q_proj", "v_proj", "k_proj", "o_proj"],
-        lora_dropout=0.05,
-        bias="none",
-        task_type="CAUSAL_LM",
-    )
-    model = get_peft_model(model, lora_config)
-
-    # Load dataset (Choose any one)
-    dataset = load_dataset("lvwerra/codeparrot-clean", split="train")  # ✅ Free dataset
-
-    # 🔥 Fix: Set `labels` properly
-    def tokenize_function(examples):
-        inputs = tokenizer(examples["content"], padding="max_length", truncation=True, max_length=512)
-        inputs["labels"] = inputs["input_ids"].copy()  # ✅ Ensure labels exist
-        return inputs
-
-    tokenized_dataset = dataset.map(tokenize_function, batched=True)
-
-    # Data collator
-    data_collator = DataCollatorForSeq2Seq(tokenizer, return_tensors="pt")
-
-    # Training arguments
-    training_args = TrainingArguments(
-        per_device_train_batch_size=1,
-        num_train_epochs=1,  # Reduce epochs for quick training
-        learning_rate=3e-4,
-        output_dir=model_path,
-        save_strategy="epoch",
-        logging_dir="./logs",
-        logging_steps=10,
-        save_total_limit=2,
-        evaluation_strategy="no",  # ✅ No eval dataset needed
-        load_best_model_at_end=False  # ✅ Prevents conflicts
-    )
-
-    # Trainer
-    trainer = Trainer(
-        model=model,
-        args=training_args,
-        train_dataset=tokenized_dataset,
-        data_collator=data_collator,
-    )
-
-    # Train
-    trainer.train()
-
-    # Save model
-    model.save_pretrained(model_path)
-    tokenizer.save_pretrained("./tokenizer")
-
-    st.write("🎉 Fine-tuning complete! Model saved.")
-
-# Chat Interface
-user_input = st.text_input("Ask a coding question:")
-if user_input:
-    inputs = tokenizer(user_input, return_tensors="pt").to("cpu")
-    outputs = model.generate(**inputs, max_length=150)
-    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    st.write("🤖 AI Tutor:", response)