auto frontend

游雁

2024-06-07 0ba1bdd476c2079f1220904d5f2a217d78bdb64a

 funasr/datasets/openai_datasets/datasets.py

@@ -180,51 +180,43 @@
        return output

    def collator(self, samples: list = None):
        outputs = {}
        for sample in samples:
            if sample is None:
                continue
            for key in sample.keys():
                if key not in outputs:
                    outputs[key] = []
                outputs[key].append(sample[key])

        for key, data_list in outputs.items():
            if isinstance(data_list[0], torch.Tensor):
                if data_list[0].dtype == torch.int64 or data_list[0].dtype == torch.int32:
        for idx in range(self.retry):
            badcase_flag = False

                    pad_value = self.int_pad_value
                else:
                    pad_value = self.float_pad_value
            outputs = {}
            for sample in samples:
                if sample is None:
                    continue
                for key in sample.keys():
                    if key not in outputs:
                        outputs[key] = []
                    outputs[key].append(sample[key])

                outputs[key] = torch.nn.utils.rnn.pad_sequence(
                    data_list, batch_first=True, padding_value=pad_value
                )

        if self.batch_type != "example":
            for i in range(10):
                outputs = self._filter_badcase(outputs, i=i)

        return outputs

    def _filter_badcase(self, outputs, i=0):
        b, t = outputs["input_ids"].shape

        if b * t > self.batch_size * 2:
            beg = torch.randint(0, 2, ()).item()
            if b < 2:
                beg = 0
            logging.info(
                f"Warning, b * t: {b * t} > {self.batch_size}, b: {b}, t: {t}, drop half data {i}th, beg:{beg}"
            )
            for key, data_list in outputs.items():
                outputs[key] = outputs[key][beg : beg + b : 2]
            #
            # speech_lengths_max = outputs["speech_lengths"].max().item()
            # outputs["speech"] = outputs["speech"][:, :speech_lengths_max, :]
            # text_lengths_max = outputs["text_lengths"].max().item()
            # outputs["text"] = outputs["text"][:, :text_lengths_max]
            # target_mask_lengths_max = outputs["target_mask_lengths"].max().item()
            # outputs["target_mask"] = outputs["target_mask"][:, :target_mask_lengths_max]
                if isinstance(data_list[0], torch.Tensor):
                    if data_list[0].dtype == torch.int64 or data_list[0].dtype == torch.int32:

                        pad_value = self.int_pad_value
                    else:
                        pad_value = self.float_pad_value

                    outputs[key] = torch.nn.utils.rnn.pad_sequence(
                        data_list, batch_first=True, padding_value=pad_value
                    )

            if self.batch_type != "example":
                b, t = outputs["input_ids"].shape
                if b * t > self.batch_size * 2:
                    beg = torch.randint(0, 2, ()).item()
                    if b < 2:
                        beg = 0
                    logging.info(
                        f"Warning, b * t: {b * t} > {self.batch_size}, b: {b}, t: {t}, drop half data {idx}th, beg:{beg}"
                    )
                    samples = samples[beg : beg + b : 2]
                    continue

            break

        return outputs