提交

kongdeqiang

2026-03-13 28ccfbfc51068a663a80764e14074df5edf2b5ba

 funasr/models/transformer/decoder.py

@@ -28,6 +28,7 @@

from funasr.register import tables


class DecoderLayer(nn.Module):
    """Single decoder layer module.

@@ -51,14 +52,14 @@
    """

    def __init__(
            self,
            size,
            self_attn,
            src_attn,
            feed_forward,
            dropout_rate,
            normalize_before=True,
            concat_after=False,
        self,
        size,
        self_attn,
        src_attn,
        feed_forward,
        dropout_rate,
        normalize_before=True,
        concat_after=False,
    ):
        """Construct an DecoderLayer object."""
        super(DecoderLayer, self).__init__()
@@ -115,9 +116,7 @@
                tgt_q_mask = tgt_mask[:, -1:, :]

        if self.concat_after:
            tgt_concat = torch.cat(
                (tgt_q, self.self_attn(tgt_q, tgt, tgt, tgt_q_mask)), dim=-1
            )
            tgt_concat = torch.cat((tgt_q, self.self_attn(tgt_q, tgt, tgt, tgt_q_mask)), dim=-1)
            x = residual + self.concat_linear1(tgt_concat)
        else:
            x = residual + self.dropout(self.self_attn(tgt_q, tgt, tgt, tgt_q_mask))
@@ -128,9 +127,7 @@
        if self.normalize_before:
            x = self.norm2(x)
        if self.concat_after:
            x_concat = torch.cat(
                (x, self.src_attn(x, memory, memory, memory_mask)), dim=-1
            )
            x_concat = torch.cat((x, self.src_attn(x, memory, memory, memory_mask)), dim=-1)
            x = residual + self.concat_linear2(x_concat)
        else:
            x = residual + self.dropout(self.src_attn(x, memory, memory, memory_mask))
@@ -146,6 +143,35 @@

        if cache is not None:
            x = torch.cat([cache, x], dim=1)

        return x, tgt_mask, memory, memory_mask


class DecoderLayerExport(nn.Module):
    def __init__(self, model):
        super().__init__()
        self.self_attn = model.self_attn
        self.src_attn = model.src_attn
        self.feed_forward = model.feed_forward
        self.norm1 = model.norm1
        self.norm2 = model.norm2
        self.norm3 = model.norm3

    def forward(self, tgt, tgt_mask, memory, memory_mask, cache=None):
        residual = tgt
        tgt = self.norm1(tgt)
        tgt_q = tgt
        tgt_q_mask = tgt_mask
        x = residual + self.self_attn(tgt_q, tgt, tgt, tgt_q_mask)

        residual = x
        x = self.norm2(x)

        x = residual + self.src_attn(x, memory, memory, memory_mask)

        residual = x
        x = self.norm3(x)
        x = residual + self.feed_forward(x)

        return x, tgt_mask, memory, memory_mask

@@ -173,15 +199,15 @@
    """

    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
        self,
        vocab_size: int,
        encoder_output_size: int,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
    ):
        super().__init__()
        attention_dim = encoder_output_size
@@ -214,11 +240,11 @@
        self.decoders = None

    def forward(
            self,
            hs_pad: torch.Tensor,
            hlens: torch.Tensor,
            ys_in_pad: torch.Tensor,
            ys_in_lens: torch.Tensor,
        self,
        hs_pad: torch.Tensor,
        hlens: torch.Tensor,
        ys_in_pad: torch.Tensor,
        ys_in_lens: torch.Tensor,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Forward decoder.

@@ -246,20 +272,14 @@
        tgt_mask = tgt_mask & m

        memory = hs_pad
        memory_mask = (~make_pad_mask(hlens, maxlen=memory.size(1)))[:, None, :].to(
            memory.device
        )
        memory_mask = (~make_pad_mask(hlens, maxlen=memory.size(1)))[:, None, :].to(memory.device)
        # Padding for Longformer
        if memory_mask.shape[-1] != memory.shape[1]:
            padlen = memory.shape[1] - memory_mask.shape[-1]
            memory_mask = torch.nn.functional.pad(
                memory_mask, (0, padlen), "constant", False
            )
            memory_mask = torch.nn.functional.pad(memory_mask, (0, padlen), "constant", False)

        x = self.embed(tgt)
        x, tgt_mask, memory, memory_mask = self.decoders(
            x, tgt_mask, memory, memory_mask
        )
        x, tgt_mask, memory, memory_mask = self.decoders(x, tgt_mask, memory, memory_mask)
        if self.normalize_before:
            x = self.after_norm(x)
        if self.output_layer is not None:
@@ -269,11 +289,11 @@
        return x, olens

    def forward_one_step(
            self,
            tgt: torch.Tensor,
            tgt_mask: torch.Tensor,
            memory: torch.Tensor,
            cache: List[torch.Tensor] = None,
        self,
        tgt: torch.Tensor,
        tgt_mask: torch.Tensor,
        memory: torch.Tensor,
        cache: List[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, List[torch.Tensor]]:
        """Forward one step.

@@ -293,9 +313,7 @@
            cache = [None] * len(self.decoders)
        new_cache = []
        for c, decoder in zip(cache, self.decoders):
            x, tgt_mask, memory, memory_mask = decoder(
                x, tgt_mask, memory, None, cache=c
            )
            x, tgt_mask, memory, memory_mask = decoder(x, tgt_mask, memory, None, cache=c)
            new_cache.append(x)

        if self.normalize_before:
@@ -310,13 +328,11 @@
    def score(self, ys, state, x):
        """Score."""
        ys_mask = subsequent_mask(len(ys), device=x.device).unsqueeze(0)
        logp, state = self.forward_one_step(
            ys.unsqueeze(0), ys_mask, x.unsqueeze(0), cache=state
        )
        logp, state = self.forward_one_step(ys.unsqueeze(0), ys_mask, x.unsqueeze(0), cache=state)
        return logp.squeeze(0), state

    def batch_score(
            self, ys: torch.Tensor, states: List[Any], xs: torch.Tensor
        self, ys: torch.Tensor, states: List[Any], xs: torch.Tensor
    ) -> Tuple[torch.Tensor, List[Any]]:
        """Score new token batch.

@@ -340,8 +356,7 @@
        else:
            # transpose state of [batch, layer] into [layer, batch]
            batch_state = [
                torch.stack([states[b][i] for b in range(n_batch)])
                for i in range(n_layers)
                torch.stack([states[b][i] for b in range(n_batch)]) for i in range(n_layers)
            ]

        # batch decoding
@@ -352,24 +367,25 @@
        state_list = [[states[i][b] for i in range(n_layers)] for b in range(n_batch)]
        return logp, state_list


@tables.register("decoder_classes", "TransformerDecoder")
class TransformerDecoder(BaseTransformerDecoder):
    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            attention_heads: int = 4,
            linear_units: int = 2048,
            num_blocks: int = 6,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            self_attention_dropout_rate: float = 0.0,
            src_attention_dropout_rate: float = 0.0,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
            concat_after: bool = False,
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
    ):
        super().__init__(
            vocab_size=vocab_size,
@@ -387,12 +403,8 @@
            num_blocks,
            lambda lnum: DecoderLayer(
                attention_dim,
                MultiHeadedAttention(
                    attention_heads, attention_dim, self_attention_dropout_rate
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                MultiHeadedAttention(attention_heads, attention_dim, self_attention_dropout_rate),
                MultiHeadedAttention(attention_heads, attention_dim, src_attention_dropout_rate),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,
@@ -404,24 +416,24 @@
@tables.register("decoder_classes", "LightweightConvolutionTransformerDecoder")
class LightweightConvolutionTransformerDecoder(BaseTransformerDecoder):
    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            attention_heads: int = 4,
            linear_units: int = 2048,
            num_blocks: int = 6,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            self_attention_dropout_rate: float = 0.0,
            src_attention_dropout_rate: float = 0.0,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
            concat_after: bool = False,
            conv_wshare: int = 4,
            conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
            conv_usebias: int = False,
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        conv_wshare: int = 4,
        conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
        conv_usebias: int = False,
    ):
        if len(conv_kernel_length) != num_blocks:
            raise ValueError(
@@ -452,9 +464,7 @@
                    use_kernel_mask=True,
                    use_bias=conv_usebias,
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                MultiHeadedAttention(attention_heads, attention_dim, src_attention_dropout_rate),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,
@@ -462,27 +472,28 @@
            ),
        )


@tables.register("decoder_classes", "LightweightConvolution2DTransformerDecoder")
class LightweightConvolution2DTransformerDecoder(BaseTransformerDecoder):
    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            attention_heads: int = 4,
            linear_units: int = 2048,
            num_blocks: int = 6,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            self_attention_dropout_rate: float = 0.0,
            src_attention_dropout_rate: float = 0.0,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
            concat_after: bool = False,
            conv_wshare: int = 4,
            conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
            conv_usebias: int = False,
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        conv_wshare: int = 4,
        conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
        conv_usebias: int = False,
    ):
        if len(conv_kernel_length) != num_blocks:
            raise ValueError(
@@ -513,9 +524,7 @@
                    use_kernel_mask=True,
                    use_bias=conv_usebias,
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                MultiHeadedAttention(attention_heads, attention_dim, src_attention_dropout_rate),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,
@@ -527,24 +536,24 @@
@tables.register("decoder_classes", "DynamicConvolutionTransformerDecoder")
class DynamicConvolutionTransformerDecoder(BaseTransformerDecoder):
    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            attention_heads: int = 4,
            linear_units: int = 2048,
            num_blocks: int = 6,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            self_attention_dropout_rate: float = 0.0,
            src_attention_dropout_rate: float = 0.0,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
            concat_after: bool = False,
            conv_wshare: int = 4,
            conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
            conv_usebias: int = False,
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        conv_wshare: int = 4,
        conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
        conv_usebias: int = False,
    ):
        if len(conv_kernel_length) != num_blocks:
            raise ValueError(
@@ -575,9 +584,7 @@
                    use_kernel_mask=True,
                    use_bias=conv_usebias,
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                MultiHeadedAttention(attention_heads, attention_dim, src_attention_dropout_rate),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,
@@ -585,27 +592,28 @@
            ),
        )


@tables.register("decoder_classes", "DynamicConvolution2DTransformerDecoder")
class DynamicConvolution2DTransformerDecoder(BaseTransformerDecoder):
    def __init__(
            self,
            vocab_size: int,
            encoder_output_size: int,
            attention_heads: int = 4,
            linear_units: int = 2048,
            num_blocks: int = 6,
            dropout_rate: float = 0.1,
            positional_dropout_rate: float = 0.1,
            self_attention_dropout_rate: float = 0.0,
            src_attention_dropout_rate: float = 0.0,
            input_layer: str = "embed",
            use_output_layer: bool = True,
            pos_enc_class=PositionalEncoding,
            normalize_before: bool = True,
            concat_after: bool = False,
            conv_wshare: int = 4,
            conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
            conv_usebias: int = False,
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        conv_wshare: int = 4,
        conv_kernel_length: Sequence[int] = (11, 11, 11, 11, 11, 11),
        conv_usebias: int = False,
    ):
        if len(conv_kernel_length) != num_blocks:
            raise ValueError(
@@ -636,9 +644,7 @@
                    use_kernel_mask=True,
                    use_bias=conv_usebias,
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                MultiHeadedAttention(attention_heads, attention_dim, src_attention_dropout_rate),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,