model.py

import os
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

MODEL_NAME = "bigcode/starcoderbase-1b"
HF_TOKEN = os.getenv("HUGGINGFACE_TOKEN")

# Force CPU mode
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

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    token=HF_TOKEN,
    torch_dtype=torch.float32,  # Use float32 for CPU
    trust_remote_code=True
).to(device)  # Move model explicitly to CPU

def generate_code(prompt: str, max_tokens: int = 256):
    inputs = tokenizer(
        prompt, 
        return_tensors="pt", 
        padding=True, 
        truncation=True,  # Allow truncation
        max_length=1024  # Set a maximum length explicitly
    ).to(device)

    output = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        pad_token_id=tokenizer.pad_token_id
    )
    
    return tokenizer.decode(output[0], skip_special_tokens=True)