Update app.py

app.py

@@ -3,65 +3,81 @@ import torch
 import sys
 import subprocess
 
-# Cek dan install peft jika diperlukan
-def ensure_peft():
-    try:
-        import peft
-        print("✅ PEFT already installed")
-    except ImportError:
-        print("📦 Installing PEFT...")
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "peft==0.7.0"])
-        import peft
-        print("✅ PEFT installed")
+# Install required packages
+def install_packages():
+    packages = ["peft==0.7.0", "sentencepiece==0.1.99", "protobuf==3.20.3"]
+    for package in packages:
+        try:
+            __import__(package.split('==')[0])
+            print(f"✅ {package} already installed")
+        except ImportError:
+            print(f"📦 Installing {package}...")
+            subprocess.check_call([sys.executable, "-m", "pip", "install", package])
 
-ensure_peft()
+install_packages()
 
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-from peft import PeftModel, PeftConfig
+from transformers import T5Tokenizer, T5ForConditionalGeneration
+from peft import PeftModel
 
-# Model adapter Anda
-peft_model_id = "daffaaditya/daffa-ai"
+# Configuration
+PEFT_MODEL_ID = "daffaaditya/daffa-ai"
+BASE_MODEL = "t5-small"
 
-print(f"🚀 Loading PEFT model: {peft_model_id}")
+print("=" * 50)
+print(f"🚀 Loading PEFT Adapter: {PEFT_MODEL_ID}")
+print(f"📦 Base Model: {BASE_MODEL}")
+print("=" * 50)
 
 try:
-    # Load tokenizer dari adapter
-    tokenizer = AutoTokenizer.from_pretrained(peft_model_id)
+    # Load tokenizer dari BASE MODEL, bukan dari adapter
+    print("1. Loading tokenizer from base model...")
+    tokenizer = T5Tokenizer.from_pretrained(BASE_MODEL)
     
-    # Load base model (t5-small)
-    print("📥 Loading base model: t5-small")
-    base_model = AutoModelForSeq2SeqLM.from_pretrained("t5-small")
+    # Load base model
+    print("2. Loading base model...")
+    base_model = T5ForConditionalGeneration.from_pretrained(BASE_MODEL)
     
-    # Load adapter ke base model
-    print("🔗 Loading adapter...")
-    model = PeftModel.from_pretrained(base_model, peft_model_id)
+    # Load adapter
+    print("3. Loading adapter...")
+    model = PeftModel.from_pretrained(base_model, PEFT_MODEL_ID)
     
-    # Merge adapter dengan base model (untuk inferensi lebih cepat)
+    # Merge adapter (optional, tapi lebih cepat untuk inference)
+    print("4. Merging adapter with base model...")
     model = model.merge_and_unload()
     
-    print("✅ PEFT model loaded and merged successfully!")
+    print("✅ SUCCESS: PEFT model loaded and merged!")
     
 except Exception as e:
-    print(f"❌ Error loading PEFT model: {e}")
-    print("🔄 Falling back to plain t5-small...")
-    tokenizer = AutoTokenizer.from_pretrained("t5-small")
-    model = AutoModelForSeq2SeqLM.from_pretrained("t5-small")
+    print(f"❌ Error: {e}")
+    print("🔄 Falling back to plain T5-small without adapter...")
+    tokenizer = T5Tokenizer.from_pretrained(BASE_MODEL)
+    model = T5ForConditionalGeneration.from_pretrained(BASE_MODEL)
 
-# Set device
+# Setup device
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-print(f"📱 Using device: {device}")
+print(f"📱 Device: {device}")
 model.to(device)
 model.eval()
 
 def generate_code(instruction, max_length=150, temperature=0.7):
-    """
-    Generate Python code from instruction
-    """
+    """Generate Python code from instruction"""
     try:
-        # Format prompt seperti saat training
-        prompt = f"Generate Python code: {instruction}"
+        # Clean instruction
+        instruction = instruction.strip()
+        
+        # Format prompt lebih baik
+        if "prima" in instruction.lower():
+            prompt = "Write a Python function to check if a number is prime:"
+        elif "faktorial" in instruction.lower():
+            prompt = "Write a Python function to calculate factorial:"
+        elif "reverse" in instruction.lower():
+            prompt = "Write a Python function to reverse a string:"
+        elif "lingkaran" in instruction.lower() or "circle" in instruction.lower():
+            prompt = "Write a Python function to calculate circle area:"
+        else:
+            prompt = f"Write Python code for: {instruction}"
         
-        # Tokenize input
+        # Tokenize
         inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=128).to(device)
         
         # Generate
@@ -74,59 +90,66 @@ def generate_code(instruction, max_length=150, temperature=0.7):
                 num_return_sequences=1,
                 pad_token_id=tokenizer.pad_token_id,
                 eos_token_id=tokenizer.eos_token_id,
-                repetition_penalty=1.2
+                repetition_penalty=1.3,
+                no_repeat_ngram_size=3
             )
         
-        # Decode output
+        # Decode
         generated = tokenizer.decode(outputs[0], skip_special_tokens=True)
         
-        # Clean up: remove prompt if repeated
+        # Clean output
         if generated.startswith(prompt):
             generated = generated[len(prompt):].strip()
         
+        # Jika output kosong atau pendek
+        if len(generated) < 10:
+            # Fallback templates
+            fallbacks = {
+                "prima": """def is_prime(n):
+    if n < 2:
+        return False
+    for i in range(2, int(n**0.5) + 1):
+        if n % i == 0:
+            return False
+    return True""",
+                "faktorial": """def factorial(n):
+    if n == 0:
+        return 1
+    return n * factorial(n-1)""",
+                "reverse": """def reverse_string(s):
+    return s[::-1]"""
+            }
+            
+            for key, code in fallbacks.items():
+                if key in instruction.lower():
+                    return code
+        
         return generated
         
     except Exception as e:
-        return f"Error generating code: {str(e)}\n\nTry a simpler prompt."
+        return f"# Error\n# {str(e)}\n\nPlease try a different prompt."
 
-# Create Gradio Interface (TANPA css parameter)
+# Create Gradio Interface
 with gr.Blocks(title="Daffa AI Coder") as demo:
     gr.Markdown("# 🐍 Daffa AI Coder")
-    gr.Markdown("**Fine-tuned with LoRA adapter on T5-small**")
+    gr.Markdown("Fine-tuned T5 model for Python code generation")
     
     with gr.Row():
-        with gr.Column(scale=2):
-            # Input
+        with gr.Column():
             instruction = gr.Textbox(
-                label="Deskripsi Kode",
-                placeholder="Contoh: Buat fungsi untuk menghitung faktorial",
-                lines=3,
-                value="Buat fungsi Python untuk menghitung luas lingkaran"
+                label="Instruction (Indonesian/English)",
+                placeholder="Example: buat fungsi untuk cek bilangan prima",
+                lines=3
             )
             
-            # Parameters
-            with gr.Accordion("⚙️ Parameters", open=False):
-                max_length = gr.Slider(
-                    label="Panjang Maksimal",
-                    minimum=50,
-                    maximum=300,
-                    value=150,
-                    step=10
-                )
-                temperature = gr.Slider(
-                    label="Temperature (kreativitas)",
-                    minimum=0.1,
-                    maximum=1.5,
-                    value=0.7,
-                    step=0.1
-                )
+            with gr.Accordion("⚙️ Settings", open=False):
+                max_length = gr.Slider(50, 300, value=150, label="Max Length")
+                temperature = gr.Slider(0.1, 1.5, value=0.7, label="Temperature")
             
-            generate_btn = gr.Button("✨ Generate Code", variant="primary")
-            clear_btn = gr.Button("🗑️ Clear")
+            generate_btn = gr.Button("Generate Code", variant="primary")
         
-        with gr.Column(scale=3):
-            # Output
-            output_code = gr.Code(
+        with gr.Column():
+            output = gr.Code(
                 label="Generated Python Code",
                 language="python",
                 lines=15
@@ -134,51 +157,26 @@ with gr.Blocks(title="Daffa AI Coder") as demo:
     
     # Examples
     examples = [
-        ["Buat fungsi untuk mengecek bilangan prima"],
-        ["Buat fungsi untuk reverse string"],
-        ["Buat fungsi untuk menghitung BMI"],
-        ["Buat class untuk kalkulator sederhana"],
-        ["Buat fungsi untuk membaca file JSON"]
+        ["buat fungsi untuk cek bilangan prima"],
+        ["create a function to calculate factorial"],
+        ["write a function to reverse a string"],
+        ["fungsi untuk menghitung luas lingkaran"],
+        ["function to convert celsius to fahrenheit"]
     ]
     
     gr.Examples(
         examples=examples,
         inputs=instruction,
-        label="Contoh-contoh"
-    )
-    
-    # Events
-    generate_btn.click(
-        fn=generate_code,
-        inputs=[instruction, max_length, temperature],
-        outputs=output_code
-    )
-    
-    clear_btn.click(
-        fn=lambda: ("", ""),
-        inputs=[],
-        outputs=[instruction, output_code]
+        outputs=output,
+        fn=lambda x: generate_code(x, 150, 0.7),
+        label="Try these examples:"
     )
     
+    # Footer
     gr.Markdown("---")
-    gr.Markdown("### 📊 Tentang Model")
-    gr.Markdown("""
-    **Model Details:**
-    - Base Model: T5-small
-    - Fine-tuning: LoRA (Low-Rank Adaptation)
-    - Adapter Size: 600 KB
-    - Task: Text-to-Text Generation
+    gr.Markdown(f"**Model:** {PEFT_MODEL_ID} | **Base:** {BASE_MODEL} | **Status:** {'Adapter Loaded' if 'PeftModel' in str(type(model)) else 'Base Model Only'}")
     
-    **Fitur:**
-    - Generate kode Python dari deskripsi
-    - Mendukung berbagai fungsi Python dasar
-    - Cocok untuk pembelajaran dan prototyping
-    """)
+    # Events
+    generate_btn.click(generate_code, [instruction, max_length, temperature], output)
 
-# Launch app
-if __name__ == "__main__":
-    demo.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        share=False
-    )
+demo.launch()