step end
(self, run_context)
| 121 | self.lang = lang |
| 122 | |
| 123 | def step_end(self, run_context): |
| 124 | """ |
| 125 | step end |
| 126 | """ |
| 127 | cb_params = run_context.original_args() |
| 128 | current_step = cb_params.cur_step_num + self.has_trained_step |
| 129 | if current_step % self.print_per_step != 0: |
| 130 | return |
| 131 | self.pplMetric.clear() |
| 132 | self.validation_loss.clear() |
| 133 | if self.parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL): |
| 134 | context.set_auto_parallel_context(strategy_ckpt_save_file="", |
| 135 | strategy_ckpt_load_file=self.strategy_ckpt_save_file) |
| 136 | rank_id = 0 |
| 137 | if self.parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, |
| 138 | ParallelMode.AUTO_PARALLEL, ParallelMode.DATA_PARALLEL): |
| 139 | rank_id = get_rank() |
| 140 | print("validation begin") |
| 141 | start_time = time.time() |
| 142 | out = self.model.eval(self.eval_dataset) |
| 143 | end_time = time.time() |
| 144 | eval_time = int(end_time - start_time) |
| 145 | |
| 146 | time_str = time.strftime("%Y-%m-%d %H:%M%S", time.localtime()) |
| 147 | out_str = f"{time_str} == Language {self.lang}; Rank {rank_id} Eval: {out}; eval_time: {eval_time}s" |
| 148 | print(out_str) |
| 149 | if self.summary_writer is not None: |
| 150 | print(f"writing: {out}") |
| 151 | tag = "validation_loss" if self.lang is None else f"validaton_loss/{self.lang}" |
| 152 | self.summary_writer.add_scalar( |
| 153 | tag=tag, |
| 154 | scalar_value=out['ppl'], |
| 155 | global_step=cb_params.cur_step_num + int(self.has_trained_step), |
| 156 | ) |
| 157 | context.set_auto_parallel_context(strategy_ckpt_save_file=self.strategy_ckpt_save_file, |
| 158 | strategy_ckpt_load_file=self.strategy_ckpt_load_file) |
| 159 | |
| 160 | |
| 161 | class SaveCheckpointCallback(Callback): |