Update app.py

app.py

@@ -1,7 +1,6 @@
 """
 Python Code Evaluator — SE-Group1
 COS60011 Technology Design Project
-
 Implements the 5-module architecture from the design document:
   1. UI Module            — Gradio web interface
   2. Validation & Flow    — Input validation and routing
@@ -9,20 +8,20 @@ Implements the 5-module architecture from the design document:
   4. Generation           — Gemma-4 LLM via Hugging Face
   5. Output               — Parse raw LLM output into structured results
 """
-
+ 
 import re
 import torch
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
-
+ 
 # ---------------------------------------------------------------------------
 # MODULE 4 — Generation Module (LLM setup)
 # ---------------------------------------------------------------------------
-MODEL_ID = "google/gemma-4-it"   # swap for "google/gemma-4-9b-it" if VRAM allows
-
-_tokenizer=None
-_pipe=None
-
+MODEL_ID = "google/gemma-4-it"
+ 
+_tokenizer = None
+_pipe = None
+ 
 def get_tokenizer():
     """Returns the tokenizer, loading it on first call."""
     global _tokenizer
@@ -48,7 +47,8 @@ def get_pipeline():
             do_sample=False,               # deterministic output
         )
     return _pipe
-
+ 
+ 
 # ---------------------------------------------------------------------------
 # MODULE 3 — Pre-processing Module (6-Element Framework)
 # ---------------------------------------------------------------------------
@@ -67,21 +67,17 @@ def build_prompt(description: str, code: str) -> str:
 You are an expert Python code reviewer. Your sole task is to determine whether the
 provided Python code correctly implements the behaviour described in the user's
 requirements description.
-
 ### CONTEXT
 Developers sometimes write code that does not fully satisfy the requirements they
 were given. You will analyse the semantic relationship between a natural-language
 description and a Python code snippet, then produce a structured evaluation report.
-
 ### INPUT DATA
 **Requirements Description:**
 {description.strip()}
-
 **Python Code:**
 ```python
 {code.strip()}
 ```
-
 ### TASK
 1. Read the requirements description carefully.
 2. Analyse the Python code line by line.
@@ -89,16 +85,13 @@ description and a Python code snippet, then produce a structured evaluation repo
 4. Estimate an accuracy percentage (0–100) reflecting how completely the code
    matches the description.
 5. List any specific requirements that are missing or incorrectly implemented.
-
 ### CONSTRAINTS
 - Do NOT execute the code.
 - Base your evaluation solely on static code analysis and logical reasoning.
 - Keep feedback concise, clear, and actionable.
 - Your response MUST follow the output format exactly.
-
 ### OUTPUT FORMAT
 Respond only with the following structure — no extra text before or after:
-
 RESULT: <PASS or FAIL>
 ACCURACY: <integer 0-100>%
 SUMMARY: <one sentence overall assessment>
@@ -110,8 +103,8 @@ ISSUES:
 <start_of_turn>model
 """
     return prompt
-
-
+ 
+ 
 # ---------------------------------------------------------------------------
 # MODULE 5 — Output Module
 # ---------------------------------------------------------------------------
@@ -124,16 +117,16 @@ def parse_output(raw: str) -> dict:
     # Strip any echoed prompt (model sometimes repeats <start_of_turn>)
     if "<start_of_turn>model" in raw:
         raw = raw.split("<start_of_turn>model")[-1]
-
+ 
     result_match   = re.search(r"RESULT:\s*(PASS|FAIL)", raw, re.IGNORECASE)
     accuracy_match = re.search(r"ACCURACY:\s*(\d{1,3})%?", raw, re.IGNORECASE)
     summary_match  = re.search(r"SUMMARY:\s*(.+)", raw, re.IGNORECASE)
     issues_match   = re.search(r"ISSUES:\s*([\s\S]+)", raw, re.IGNORECASE)
-
+ 
     result   = result_match.group(1).upper()  if result_match   else "UNKNOWN"
     accuracy = int(accuracy_match.group(1))   if accuracy_match else -1
     summary  = summary_match.group(1).strip() if summary_match  else "Could not extract summary."
-
+ 
     if issues_match:
         raw_issues = issues_match.group(1).strip()
         issues = [
@@ -143,7 +136,7 @@ def parse_output(raw: str) -> dict:
         ]
     else:
         issues = ["Could not extract issues from model response."]
-
+ 
     return {
         "result":   result,
         "accuracy": accuracy,
@@ -151,8 +144,8 @@ def parse_output(raw: str) -> dict:
         "issues":   issues,
         "raw":      raw.strip(),
     }
-
-
+ 
+ 
 def format_for_display(parsed: dict) -> tuple[str, str, str]:
     """
     Converts the parsed dict into three Gradio-friendly strings:
@@ -160,24 +153,24 @@ def format_for_display(parsed: dict) -> tuple[str, str, str]:
       - metrics     (accuracy + summary)
       - issues_text (bullet list)
     """
-    emoji  = "✅" if parsed["result"] == "PASS" else ("❌" if parsed["result"] == "FAIL" else "⚠️")
+    emoji   = "✅" if parsed["result"] == "PASS" else ("❌" if parsed["result"] == "FAIL" else "⚠️")
     verdict = f"{emoji}  {parsed['result']}"
-
+ 
     acc_str = f"{parsed['accuracy']}%" if parsed["accuracy"] >= 0 else "N/A"
     metrics = f"Accuracy: {acc_str}\n\nSummary: {parsed['summary']}"
-
+ 
     issues_text = "\n".join(f"• {issue}" for issue in parsed["issues"])
     return verdict, metrics, issues_text
-
-
+ 
+ 
 # ---------------------------------------------------------------------------
 # MODULE 4 — Generation Module (inference call)
 # ---------------------------------------------------------------------------
 def generate(prompt: str) -> str:
-    outputs = pipe(prompt, return_full_text=False)
+    outputs = get_pipeline()(prompt, return_full_text=False)  # FIX: was pipe() — undefined variable
     return outputs[0]["generated_text"]
-
-
+ 
+ 
 # ---------------------------------------------------------------------------
 # MODULE 2 — Validation & Flow Management Module
 # ---------------------------------------------------------------------------
@@ -191,48 +184,45 @@ def validate_and_evaluate(description: str, code: str):
         return "", "", "", "⚠️ Please provide a requirements description."
     if not code or not code.strip():
         return "", "", "", "⚠️ Please provide Python code to evaluate."
-
+ 
     # --- Pre-processing ---
     prompt = build_prompt(description, code)
-
+ 
+    # Token length guard — warn user before sending oversized input to the model
     token_count = len(get_tokenizer().encode(prompt))
     if token_count > 2048:
         return "", "", "", (
             f"⚠️ Input too long ({token_count} tokens). "
             "Please shorten your description or code and try again."
         )
-    
+ 
     # --- Generation ---
     try:
         raw_output = generate(prompt)
     except Exception as exc:
         return "", "", "", f"❌ Model error: {exc}"
-
+ 
     # --- Output ---
     parsed  = parse_output(raw_output)
     verdict, metrics, issues = format_for_display(parsed)
-
+ 
     return verdict, metrics, issues, ""   # empty error = success
-
-
+ 
+ 
 # ---------------------------------------------------------------------------
 # MODULE 1 — UI Module (Gradio)
 # ---------------------------------------------------------------------------
-with gr.Blocks(
-    title="Python Code Evaluator — SE-Group1",
-    theme=gr.themes.Soft(primary_hue="blue"),
-) as demo:
-
+with gr.Blocks(title="Python Code Evaluator — SE-Group1") as demo:  # FIX: theme moved to launch()
+ 
     gr.Markdown(
         """
         # 🐍 Python Code Evaluator
         **COS60011 — SE-Group1** | Powered by Gemma-4 via Hugging Face
-
         Enter a **requirements description** and your **Python code**.
         The system will evaluate whether the code correctly implements the described behaviour.
         """
     )
-
+ 
     with gr.Row():
         with gr.Column(scale=1):
             description_input = gr.Textbox(
@@ -247,19 +237,19 @@ with gr.Blocks(
                 value='def add(a, b):\n    return a + b\n',
             )
             submit_btn = gr.Button("▶  Evaluate", variant="primary", size="lg")
-
+ 
         with gr.Column(scale=1):
-            error_output  = gr.Textbox(label="⚠️ Validation Error", visible=True, interactive=False)
+            error_output   = gr.Textbox(label="⚠️ Validation Error", visible=True, interactive=False)
             verdict_output = gr.Textbox(label="🏁 Verdict",  interactive=False, lines=1)
             metrics_output = gr.Textbox(label="📊 Metrics & Summary", interactive=False, lines=4)
             issues_output  = gr.Textbox(label="🔍 Issues Found", interactive=False, lines=8)
-
+ 
     submit_btn.click(
         fn=validate_and_evaluate,
         inputs=[description_input, code_input],
         outputs=[verdict_output, metrics_output, issues_output, error_output],
     )
-
+ 
     gr.Markdown(
         """
         ---
@@ -267,6 +257,6 @@ with gr.Blocks(
         > Results should be treated as supplementary feedback, not a replacement for unit testing.
         """
     )
-
+ 
 if __name__ == "__main__":
-    demo.launch(share=False)
+    demo.launch(share=False, theme=gr.themes.Soft(primary_hue="blue"))