Create app.py

app.py

@@ -0,0 +1,114 @@
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from peft import PeftModel
+
+# --- Configuration ---
+# 1. Base Model ID: Llama-2-7b-chat-hf is typically used as the base
+base_model_id = "meta-llama/Llama-2-7b-chat-hf"
+
+# 2. LoRA Path: IMPORTANT! Replace this with the path to your fine-tuned model
+# This should be the Hugging Face repo ID (e.g., "your-username/llama2-dockerfile-lora")
+# or a local directory path where the adapter weights are stored.
+lora_path = "Arsh014/lora-llama2-finetuned"
+
+# Check for CUDA availability
+device = 0 if torch.cuda.is_available() else -1
+
+print(f"Loading tokenizer from: {base_model_id}")
+tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+
+# 3. Load the base model with 8-bit quantization for efficiency
+print(f"Loading base model (8-bit) from: {base_model_id}")
+model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    load_in_8bit=True,
+    torch_dtype=torch.float16,
+    device_map="auto"
+)
+
+# 4. Apply the PEFT (LoRA) adapters to the base model
+print(f"Applying LoRA adapter from: {lora_path}")
+try:
+    model = PeftModel.from_pretrained(model, lora_path)
+    model.eval() # Set model to evaluation mode
+except Exception as e:
+    print(f"Error loading LoRA adapter from {lora_path}. Ensure it exists and is correct.")
+    print(f"Error: {e}")
+    # The app will likely fail if the LoRA path is incorrect.
+    # We proceed with the base model, but generation quality will be poor for the task.
+
+# 5. Create a text-generation pipeline
+print("Creating text-generation pipeline.")
+pipe = pipeline(
+    "text-generation", 
+    model=model, 
+    tokenizer=tokenizer, 
+    device=device
+)
+
+def format_prompt(instruction, code):
+    """Formats the instruction and input code into the required prompt template."""
+    return f"""### Instruction:
+{instruction}
+
+### Input:
+{code}
+
+### Response:"""
+
+def explain_dockerfile(instruction, code):
+    """Generates the explanation using the text-generation pipeline."""
+    if not instruction or not code:
+        return "Please provide both an instruction and the Dockerfile code."
+
+    prompt = format_prompt(instruction, code)
+
+    # Generate response
+    response = pipe(
+        prompt, 
+        max_new_tokens=256, 
+        do_sample=True, 
+        temperature=0.7,
+        return_full_text=False # We want only the new tokens generated after the prompt
+    )
+    
+    # The pipeline's output can be complex, extract the text and clean up
+    generated_text = response[0]["generated_text"].strip()
+    
+    # Clean up the output to remove the initial prompt if return_full_text=False 
+    # didn't perfectly handle it (it's good practice to split/strip again)
+    if "### Response:" in generated_text:
+        return generated_text.split("### Response:")[-1].strip()
+    
+    return generated_text
+
+# 6. Gradio Interface
+print("Launching Gradio Interface...")
+iface = gr.Interface(
+    fn=explain_dockerfile,
+    inputs=[
+        gr.Textbox(
+            label="Instruction", 
+            placeholder="e.g., Explain the function of each line and the overall goal of this Dockerfile.",
+            value="Explain this Dockerfile in detail and suggest one security improvement.",
+            lines=2
+        ),
+        gr.Textbox(
+            label="Dockerfile Code", 
+            lines=10, 
+            placeholder="Paste your Dockerfile here, e.g., \nFROM python:3.9-slim\nWORKDIR /app\nCOPY requirements.txt .\nRUN pip install -r requirements.txt\nCOPY . .\nCMD [\"python\", \"app.py\"]",
+            value="FROM node:18-alpine\nWORKDIR /usr/src/app\nCOPY package*.json ./ \nRUN npm install\nCOPY . .\nEXPOSE 3000\nCMD [ \"npm\", \"start\" ]"
+        )
+    ],
+    outputs=gr.Textbox(
+        label="Explanation (Generated by LoRA Model)", 
+        lines=15
+    ),
+    title="LoRA-Tuned Llama-2 Dockerfile Explainer",
+    description="A simple application to explain complex Dockerfiles using a fine-tuned Llama-2 model (via LoRA).",
+    live=False
+)
+
+if __name__ == "__main__":
+    iface.launch()