"""Minimal local generation example for QuadOrbit-40M."""

from pathlib import Path

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer


model_dir = Path(__file__).resolve().parent
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if device == "cuda" else torch.float32

tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_dir,
    trust_remote_code=True,
    torch_dtype=dtype,
).to(device)

inputs = tokenizer("The future of language models", return_tensors="pt").to(device)
with torch.no_grad():
    generated = model.generate(
        **inputs,
        max_new_tokens=30,
        do_sample=True,
        temperature=0.8,
        top_k=50,
        use_cache=False,
    )
print(tokenizer.decode(generated[0], skip_special_tokens=True))