| 1358 | return total_latency, total_breakdown |
| 1359 | |
| 1360 | def print_config(self, name="Training Configs") -> None: |
| 1361 | config_str = f"\n{name.center(PRINT_LINE_WIDTH, '-')}\n" |
| 1362 | config_str += f"{pformat(self.model_config)}\n" |
| 1363 | config_str += f"{pformat(self.gpu_config)}\n" |
| 1364 | config_str += f"{pformat(self.dtype_config)}\n" |
| 1365 | config_str += f"{pformat(self.parallelism_config)}\n" |
| 1366 | logger.info(config_str) |
| 1367 | |
| 1368 | def get_configs_desc(self) -> str: |
| 1369 | return f"{self.model_config.name}-{self.gpu_config.name}-{self.dtype_config.name}-tp{self.parallelism_config.tp_size}-pp{self.parallelism_config.pp_size}-dp{self.parallelism_config.dp_size}-sp{self.parallelism_config.sp_size}-fe{round(self.flops_efficiency, 2)}-ep{self.parallelism_config.ep_size}-hbme{round(self.hbm_memory_efficiency, 2)}" |