updated model.py

model.py

@@ -10,40 +10,39 @@ device = "cpu"
 # Load tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)
 
-# Ensure the tokenizer has a pad token set
 if tokenizer.pad_token is None:
-    tokenizer.pad_token = tokenizer.eos_token  # Set pad_token to eos_token
+    tokenizer.pad_token = tokenizer.eos_token  
 
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
     token=HF_TOKEN,
-    torch_dtype=torch.float32,  # Ensure compatibility with CPU
+    torch_dtype=torch.float32,  
     trust_remote_code=True
 ).to(device)
 
 def generate_code(prompt: str, max_tokens: int = 256):
-    formatted_prompt = f"# Python\n{prompt}\n\n"  # Ensure the model understands it's code
+    # 🛠 **Improve Speed & Support Multi-language**
+    formatted_prompt = f"{prompt}\n### Code:\n"  # Hint that code follows
 
     inputs = tokenizer(
         formatted_prompt, 
         return_tensors="pt", 
-        padding=True, 
         truncation=True,  
-        max_length=1024  # Explicit max length to prevent issues
+        max_length=512  # ⏩ Reduce max length to speed up processing
     ).to(device)
 
     output = model.generate(
         **inputs,
         max_new_tokens=max_tokens,
         pad_token_id=tokenizer.pad_token_id,
-        do_sample=True,  # Enable randomness for better outputs
-        top_p=0.95,  # Nucleus sampling to improve generation
-        temperature=0.7  # Control creativity
+        do_sample=True,  
+        top_p=0.90,  # 🎯 Prioritize better predictions
+        temperature=0.6  # 🔥 More deterministic output
     )
     
     generated_code = tokenizer.decode(output[0], skip_special_tokens=True)
 
-    # Clean the output: remove the repeated prompt at the start
+    # Remove the input prompt from the output
     if generated_code.startswith(formatted_prompt):
         generated_code = generated_code[len(formatted_prompt):]