t0 = time.perf_counter()

spec_out = target.generate(
    **inputs, max_new_tokens=200, do_sample=False,
    assistant_model=draft,
    streamer=streamer,
)

spec_time = time.perf_counter() - t0

print(f"\n⏱ With Speculative Decoding:")
print(f"  Runtime: {spec_time:.2f}s")
print(f"  Token speed: {len(spec_out[0]) / spec_time:.1f} tok/s")
print(f"  Speedup over standard decoding: {base_time / spec_time:.2f}x")