from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch

MODEL_REPO = "Rahul-8799/software_architect_command_r"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4"
)

tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_REPO,
    trust_remote_code=True,
    quantization_config=bnb_config,
    device_map={"": 0},  # force full GPU load
    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)