Update app.py

app.py

@@ -1,25 +1,24 @@
 import os
 from flask import Flask, request, jsonify
-from transformers import RobertaTokenizer, T5ForConditionalGeneration
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch
 
 app = Flask(__name__)
 
-# --- CRITICAL CHANGE ---
-# Instead of your local folder, we point to a PUBLIC Expert Model
-# This model has read millions of commit messages and knows exactly what to do.
-MODEL_NAME = "ncoop57/commit-t5"
+# --- UPDATED MODEL ---
+# This model is specifically trained for git commit generation and is active.
+MODEL_NAME = "SEBIS/code_trans_t5_base_commit_generation"
 
 print(f"--- AI Commit Generator Server ---")
-print(f"Downloading/Loading Expert Model: {MODEL_NAME}")
+print(f"Downloading/Loading Model: {MODEL_NAME}")
 
-device = "cpu" # HF Spaces free tier is CPU
+device = "cpu" 
 
 try:
-    # This will download the model automatically the first time it runs
-    tokenizer = RobertaTokenizer.from_pretrained(MODEL_NAME)
-    model = T5ForConditionalGeneration.from_pretrained(MODEL_NAME).to(device)
-    print("✅ Expert Model loaded successfully!")
+    # Use AutoTokenizer and AutoModelForSeq2SeqLM for better compatibility
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, skip_special_tokens=True)
+    model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME).to(device)
+    print("✅ Model loaded successfully!")
 except Exception as e:
     print(f"❌ Error loading model: {e}")
     exit(1)
@@ -28,19 +27,14 @@ def generate_summary(diff_text):
     if not diff_text or len(diff_text.strip()) < 5:
         return "Update file"
 
-    # Preprocess: "commit-t5" just expects the raw code diff, no "Summarize:" prefix needed usually,
-    # but let's keep it simple.
-    input_text = diff_text + " </s>" 
-    
-    input_ids = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True).input_ids.to(device)
+    # This model works best with raw code, but we tokenize it first
+    input_ids = tokenizer.encode(diff_text, return_tensors="pt", max_length=512, truncation=True).to(device)
 
     outputs = model.generate(
         input_ids, 
-        max_length=80,      # Commits are usually short
-        min_length=5,
+        max_length=80,
         num_beams=5, 
-        early_stopping=True,
-        no_repeat_ngram_size=2 # Stops it from saying "update update update"
+        early_stopping=True
     )
     return tokenizer.decode(outputs[0], skip_special_tokens=True)
 
@@ -58,14 +52,13 @@ def generate_commit():
         name = file_obj.get('name', 'Unknown File')
         diff = file_obj.get('diff', '')
         
-        # Skip binary files or huge diffs
-        if len(diff) > 4000:
+        # Skip huge files to prevent crashing CPU
+        if len(diff) > 6000:
             final_message_parts.append(f"{name}\nLarge changes detected")
             continue
 
         try:
             summary = generate_summary(diff)
-            # Format: File Name -> The generated message
             final_message_parts.append(f"{name}\n{summary}")
         except Exception as e:
             print(f"Error processing {name}: {e}")