Tok/sec calc in demo.py should not include TTFT

[mattjcly]

mlx-engine/demo.py

Lines 114 to 122 in cb1b880

	def print_stats(self):
	"""Print generation statistics."""
	end_time = time.time()
	total_time = end_time - self.start_time
	print(f"\n\nGeneration stats:")
	print(f" - Time to first token: {self.first_token_time - self.start_time:.2f}s")
	print(f" - Total tokens generated: {self.total_tokens}")
	print(f" - Total time: {total_time:.2f}s")
	print(f" - Tokens per second: {self.total_tokens / total_time:.2f}")

Tokens per second should be calculated by dividing by total_time - time_to_first_token instead of just total_time