Dev gzf deepspeed (#1844)

zhifu gao

2024-06-24 abb33d6b2097e5b0643326bc1b376a63cdc2f967

 funasr/models/sense_voice/model.py

@@ -10,12 +10,13 @@
from torch import Tensor
from torch import nn
from torch.cuda.amp import autocast
from funasr.metrics.compute_acc import compute_accuracy
from funasr.metrics.compute_acc import compute_accuracy, th_accuracy
from funasr.losses.label_smoothing_loss import LabelSmoothingLoss
from funasr.train_utils.device_funcs import force_gatherable
from . import whisper_lib as whisper
from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
from funasr.utils.datadir_writer import DatadirWriter
from funasr.models.ctc.ctc import CTC

from funasr.register import tables

@@ -73,8 +74,6 @@
    ):
        target_mask = kwargs.get("target_mask", None)

        # import pdb;
        # pdb.set_trace()
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
@@ -303,8 +302,6 @@
    ):
        target_mask = kwargs.get("target_mask", None)

        # import pdb;
        # pdb.set_trace()
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
@@ -648,8 +645,6 @@
    ):
        target_mask = kwargs.get("target_mask", None)

        # import pdb;
        # pdb.set_trace()
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
@@ -667,9 +662,11 @@
        else:
            encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)

        loss_att, acc_att, cer_att, wer_att = self._calc_att_loss(
            encoder_out, encoder_out_lens, text, text_lengths, target_mask=target_mask
        )
        with autocast(False):
            loss_att, acc_att, cer_att, wer_att = self._calc_att_loss(
                encoder_out, encoder_out_lens, text, text_lengths, target_mask=target_mask
            )

        loss = loss_att
        stats = {}
        stats["acc"] = acc_att
@@ -1041,6 +1038,7 @@
        self.length_normalized_loss = length_normalized_loss
        self.beam_search = None
        self.activation_checkpoint = kwargs.get("activation_checkpoint", False)
        self.encoder_output_size = encoder_output_size

    def forward(
        self,
@@ -1052,8 +1050,6 @@
    ):
        target_mask = kwargs.get("target_mask", None)

        # import pdb;
        # pdb.set_trace()
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
@@ -1264,15 +1260,31 @@
        if isinstance(task, str):
            task = [task]
        task = "".join([f"<|{x}|>" for x in task])
        initial_prompt = kwargs.get("initial_prompt", f"<|startoftranscript|>{task}")

        language = DecodingOptions.get("language", None)
        language = None if language == "auto" else language
        sos = kwargs.get("model_conf").get("sos")
        if isinstance(sos, str):
            initial_prompt = kwargs.get("initial_prompt", f"<|startoftranscript|>{task}")

        sos = f"{initial_prompt}<|{language}|>" if language is not None else initial_prompt
        sos_int = tokenizer.encode(sos, allowed_special="all")
            language = DecodingOptions.get("language", None)
            language = None if language == "auto" else language

            sos = f"{initial_prompt}<|{language}|>" if language is not None else initial_prompt
            sos_int = tokenizer.encode(sos, allowed_special="all")
        else:
            language = DecodingOptions.get("language", None)
            language = None if language == "auto" else language
            initial_prompt = kwargs.get("initial_prompt", f"{task}")
            initial_prompt_lid = (
                f"{initial_prompt}<|{language}|>" if language is not None else initial_prompt
            )
            initial_prompt_lid_int = tokenizer.encode(initial_prompt_lid, allowed_special="all")
            sos_int = [sos] + initial_prompt_lid_int
        eos = kwargs.get("model_conf").get("eos")
        eos_int = tokenizer.encode(eos, allowed_special="all")
        if isinstance(eos, str):
            eos_int = tokenizer.encode(eos, allowed_special="all")
        else:
            eos_int = [eos]

        self.beam_search.sos = sos_int
        self.beam_search.eos = eos_int[0]

@@ -1297,9 +1309,7 @@
        )
        self.beam_search.event_score_ga = DecodingOptions.get("gain_tokens_score", [1, 1, 1, 1])

        encoder_out, encoder_out_lens = self.encode(
            speech[None, :, :].permute(0, 2, 1), speech_lengths
        )
        encoder_out, encoder_out_lens = self.encode(speech[None, :, :], speech_lengths)

        if text_token_int is not None:
            i = 0
@@ -1378,3 +1388,7 @@
                    ibest_writer["text"][key[i]] = text

        return results, meta_data


from funasr.models.paraformer.search import Hypothesis
from funasr.utils import postprocess_utils