def run()

in benchmark.py [0:0]

• 53 lines of code
• 10 McCabe index (conditional complexity)

    def run(self):
        def _step():
            t_step_start = self.time_fn()
            with self.amp_autocast():
                output = self.model(self.example_inputs)
            t_step_end = self.time_fn(True)
            return t_step_end - t_step_start

        _logger.info(
            f'Running inference benchmark on {self.model_name} for {self.num_bench_iter} steps w/ '
            f'input size {self.input_size} and batch size {self.batch_size}.')

        with torch.no_grad():
            self._init_input()

            for _ in range(self.num_warm_iter):
                _step()

            total_step = 0.
            num_samples = 0
            t_run_start = self.time_fn()
            for i in range(self.num_bench_iter):
                delta_fwd = _step()
                total_step += delta_fwd
                num_samples += self.batch_size
                num_steps = i + 1
                if num_steps % self.log_freq == 0:
                    _logger.info(
                        f"Infer [{num_steps}/{self.num_bench_iter}]."
                        f" {num_samples / total_step:0.2f} samples/sec."
                        f" {1000 * total_step / num_steps:0.3f} ms/step.")
            t_run_end = self.time_fn(True)
            t_run_elapsed = t_run_end - t_run_start

        results = dict(
            samples_per_sec=round(num_samples / t_run_elapsed, 2),
            step_time=round(1000 * total_step / self.num_bench_iter, 3),
            batch_size=self.batch_size,
            img_size=self.input_size[-1],
            param_count=round(self.param_count / 1e6, 2),
        )

        retries = 0 if self.compiled else 2  # skip profiling if model is scripted
        while retries:
            retries -= 1
            try:
                if has_deepspeed_profiling:
                    macs, _ = profile_deepspeed(self.model, self.input_size)
                    results['gmacs'] = round(macs / 1e9, 2)
                elif has_fvcore_profiling:
                    macs, activations = profile_fvcore(self.model, self.input_size, force_cpu=not retries)
                    results['gmacs'] = round(macs / 1e9, 2)
                    results['macts'] = round(activations / 1e6, 2)
            except RuntimeError as e:
                pass

        _logger.info(
            f"Inference benchmark of {self.model_name} done. "
            f"{results['samples_per_sec']:.2f} samples/sec, {results['step_time']:.2f} ms/step")

        return results