rag-end2end-retriever/finetune_rag.py [414:487]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def save_metrics(self, latest_metrics, type_path) -> None:
        self.metrics[type_path].append(latest_metrics)
        save_json(self.metrics, self.metrics_save_path)

    def calc_generative_metrics(self, preds, target) -> Dict:
        return calculate_exact_match(preds, target)

    def _generative_step(self, batch: dict) -> dict:
        start_time = time.time()
        batch = BatchEncoding(batch).to(device=self.model.device)
        generated_ids = self.model.generate(
            batch["input_ids"],
            attention_mask=batch["attention_mask"],
            do_deduplication=False,  # rag specific parameter
            use_cache=True,
            min_length=1,
            max_length=self.target_lens["val"],
        )
        gen_time = (time.time() - start_time) / batch["input_ids"].shape[0]
        preds: List[str] = self.ids_to_clean_text(generated_ids)
        target: List[str] = self.ids_to_clean_text(batch["decoder_input_ids"])
        # print(preds,target)
        loss_tensors = self._step(batch)
        base_metrics = dict(zip(self.loss_names, loss_tensors))
        gen_metrics: Dict = self.calc_generative_metrics(preds, target)

        summ_len = np.mean(lmap(len, generated_ids))
        base_metrics.update(gen_time=gen_time, gen_len=summ_len, preds=preds, target=target, **gen_metrics)
        return base_metrics

    def test_step(self, batch, batch_idx):
        return self._generative_step(batch)

    def test_epoch_end(self, outputs):
        return self.validation_epoch_end(outputs, prefix="test")

    def get_dataset(self, type_path) -> Seq2SeqDataset:
        n_obs = self.n_obs[type_path]
        max_target_length = self.target_lens[type_path]
        dataset = Seq2SeqDataset(
            self.tokenizer,
            type_path=type_path,
            n_obs=n_obs,
            max_target_length=max_target_length,
            **self.dataset_kwargs,
        )
        return dataset

    def get_dataloader(self, type_path: str, batch_size: int, shuffle: bool = False) -> DataLoader:
        dataset = self.get_dataset(type_path)

        dataloader = DataLoader(
            dataset,
            batch_size=batch_size,
            collate_fn=dataset.collate_fn,
            shuffle=shuffle,
            num_workers=self.num_workers,
        )
        return dataloader

    def train_dataloader(self) -> DataLoader:
        dataloader = self.get_dataloader("train", batch_size=self.hparams.train_batch_size, shuffle=True)
        return dataloader

    def val_dataloader(self) -> DataLoader:
        return self.get_dataloader("val", batch_size=self.hparams.eval_batch_size)

    def test_dataloader(self) -> DataLoader:
        return self.get_dataloader("test", batch_size=self.hparams.eval_batch_size)

    @pl.utilities.rank_zero_only
    def on_save_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
        save_path = self.output_dir.joinpath("checkpoint{}".format(self.step_count))
        self.model.config.save_step = self.step_count
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



rag/finetune_rag.py [301:374]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def save_metrics(self, latest_metrics, type_path) -> None:
        self.metrics[type_path].append(latest_metrics)
        save_json(self.metrics, self.metrics_save_path)

    def calc_generative_metrics(self, preds, target) -> Dict:
        return calculate_exact_match(preds, target)

    def _generative_step(self, batch: dict) -> dict:
        start_time = time.time()
        batch = BatchEncoding(batch).to(device=self.model.device)
        generated_ids = self.model.generate(
            batch["input_ids"],
            attention_mask=batch["attention_mask"],
            do_deduplication=False,  # rag specific parameter
            use_cache=True,
            min_length=1,
            max_length=self.target_lens["val"],
        )

        gen_time = (time.time() - start_time) / batch["input_ids"].shape[0]
        preds: List[str] = self.ids_to_clean_text(generated_ids)
        target: List[str] = self.ids_to_clean_text(batch["decoder_input_ids"])
        loss_tensors = self._step(batch)
        base_metrics = dict(zip(self.loss_names, loss_tensors))
        gen_metrics: Dict = self.calc_generative_metrics(preds, target)

        summ_len = np.mean(lmap(len, generated_ids))
        base_metrics.update(gen_time=gen_time, gen_len=summ_len, preds=preds, target=target, **gen_metrics)
        return base_metrics

    def test_step(self, batch, batch_idx):
        return self._generative_step(batch)

    def test_epoch_end(self, outputs):
        return self.validation_epoch_end(outputs, prefix="test")

    def get_dataset(self, type_path) -> Seq2SeqDataset:
        n_obs = self.n_obs[type_path]
        max_target_length = self.target_lens[type_path]
        dataset = Seq2SeqDataset(
            self.tokenizer,
            type_path=type_path,
            n_obs=n_obs,
            max_target_length=max_target_length,
            **self.dataset_kwargs,
        )
        return dataset

    def get_dataloader(self, type_path: str, batch_size: int, shuffle: bool = False) -> DataLoader:
        dataset = self.get_dataset(type_path)

        dataloader = DataLoader(
            dataset,
            batch_size=batch_size,
            collate_fn=dataset.collate_fn,
            shuffle=shuffle,
            num_workers=self.num_workers,
        )
        return dataloader

    def train_dataloader(self) -> DataLoader:
        dataloader = self.get_dataloader("train", batch_size=self.hparams.train_batch_size, shuffle=True)
        return dataloader

    def val_dataloader(self) -> DataLoader:
        return self.get_dataloader("val", batch_size=self.hparams.eval_batch_size)

    def test_dataloader(self) -> DataLoader:
        return self.get_dataloader("test", batch_size=self.hparams.eval_batch_size)

    @pl.utilities.rank_zero_only
    def on_save_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
        save_path = self.output_dir.joinpath("checkpoint{}".format(self.step_count))
        self.model.config.save_step = self.step_count
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -