model.py

# import os
# from transformers import AutoModelForCausalLM, AutoTokenizer
# import torch

# # Correct model name
# MODEL_NAME = "bigcode/starcoder"

# # Ensure the token is provided
# HF_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
# if not HF_TOKEN:
#     raise ValueError("Missing Hugging Face token. Set HUGGINGFACE_TOKEN as an environment variable.")

# # Set device
# device = "cuda" if torch.cuda.is_available() else "cpu"

# # Load tokenizer with authentication
# tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)

# # Load model with optimizations
# model = AutoModelForCausalLM.from_pretrained(
#     MODEL_NAME,
#     token=HF_TOKEN,
#     torch_dtype=torch.float16,  # Reduce memory usage
#     low_cpu_mem_usage=True,     # Optimize loading
#     device_map="auto",         # Automatic device placement
#     offload_folder="offload"    # Offload to disk if needed
# ).to(device)

# def generate_code(prompt: str, max_tokens: int = 256):
#     """Generates code based on the input prompt."""
#     if not prompt.strip():
#         return "Error: Empty prompt provided."

#     inputs = tokenizer(prompt, return_tensors="pt").to(device)
#     output = model.generate(**inputs, max_new_tokens=max_tokens)
#     return tokenizer.decode(output[0], skip_special_tokens=True)

import os
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

MODEL_NAME = "bigcode/starcoderbase-3b"  # Lighter version
HF_TOKEN = os.getenv("HUGGINGFACE_TOKEN")

quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16
)

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    token=HF_TOKEN,
    quantization_config=quant_config,
    device_map="auto",
    trust_remote_code=True
)

def generate_code(prompt: str, max_tokens: int = 256):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    output = model.generate(**inputs, max_new_tokens=max_tokens)
    return tokenizer.decode(output[0], skip_special_tokens=True)