from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

MODEL_REPO = "Rahul-8799/software_engineer_mellum"

tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
model = AutoModelForCausalLM.from_pretrained(MODEL_REPO, device_map="auto", torch_dtype=torch.float16)
model.eval()

def run(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(**inputs, max_new_tokens=512)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)