Upload 3 files

app.py

@@ -0,0 +1,24 @@
+# CODEXGAME/backend/ai_evaluator/app.py
+
+import gradio as gr
+from tinyllama_inference import evaluate_code
+
+def evaluate_interface(question, code):
+    result = evaluate_code(question, code)
+    stars = result.get("stars", 0)
+    feedback = result.get("feedback", "No feedback provided.")
+    return f"Stars: {stars}\nFeedback: {feedback}"
+
+iface = gr.Interface(
+    fn=evaluate_interface,
+    inputs=[
+        gr.inputs.Textbox(lines=2, placeholder="Enter the problem question here..."),
+        gr.inputs.Textbox(lines=10, placeholder="Enter your code solution here...")
+    ],
+    outputs="text",
+    title="TinyLlama Code Evaluator",
+    description="Evaluate your coding solution with TinyLlama. Provide the problem statement and your code to get a rating and feedback."
+)
+
+if __name__ == "__main__":
+    iface.launch()

requirements.txt

@@ -0,0 +1,4 @@
+torch
+transformers
+accelerate
+gradio

tinyllama_inference.py

@@ -0,0 +1,44 @@
+# CODEXGAME/backend/ai_evaluator/tinyllama_inference.py
+
+import json
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+def load_model():
+    # Change the model identifier if needed – this should be a TinyLlama variant available on Hugging Face.
+    model_name = "TheBloke/tiny-llama-7b"
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForCausalLM.from_pretrained(model_name)
+    return tokenizer, model
+
+def evaluate_code(question, code):
+    # Construct a prompt for the AI evaluator.
+    prompt = f"""
+You are an expert code evaluator.
+Rate the user's solution to the following problem from 0-5 (0 = completely incorrect, 5 = excellent).
+Also provide a concise "feedback" message.
+Problem: "{question}"
+Solution: "{code}"
+Return ONLY valid JSON: {{"stars": number, "feedback": string}}
+Do not include any extra text outside the JSON.
+"""
+    tokenizer, model = load_model()
+    inputs = tokenizer(prompt, return_tensors="pt")
+    outputs = model.generate(**inputs, max_new_tokens=150)
+    response_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    try:
+        result = json.loads(response_text.strip())
+    except Exception as e:
+        result = {"stars": 0, "feedback": "Evaluation failed. Unable to parse AI response."}
+    return result
+
+# For direct testing from the command line
+if __name__ == "__main__":
+    import sys
+    if len(sys.argv) < 3:
+        print(json.dumps({"error": "Please provide a question and code as arguments"}))
+        sys.exit(1)
+    question = sys.argv[1]
+    code = sys.argv[2]
+    result = evaluate_code(question, code)
+    print(json.dumps(result))