Dev gzf exp (#1665)

zhifu gao

2024-04-26 1cdb3cc28d4d89a576cc06e5cd8eb80da1f3a3aa

 funasr/models/conformer_rwkv/decoder.py

@@ -28,6 +28,12 @@
from omegaconf import OmegaConf
from funasr.register import tables


class LayerNorm(nn.LayerNorm):
    def forward(self, x):
        return super().forward(x.float()).type(x.dtype)


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

@@ -51,21 +57,22 @@
    """

    def __init__(
            self,
            size,
            self_attn,
            src_attn,
            feed_forward,
            dropout_rate,
            normalize_before=True,
            concat_after=False,
            layer_id=None,
            args={},
        self,
        size,
        # self_attn,
        src_attn,
        feed_forward,
        dropout_rate,
        normalize_before=True,
        concat_after=False,
        layer_id=None,
        args={},
        **kwargs,
    ):
        """Construct an DecoderLayer object."""
        super(DecoderLayer, self).__init__()
        self.size = size
        self.self_attn = self_attn.to(torch.bfloat16)
        # self.self_attn = self_attn.to(torch.bfloat16)
        self.src_attn = src_attn
        self.feed_forward = feed_forward
        self.norm1 = LayerNorm(size)
@@ -78,6 +85,20 @@
            self.concat_linear1 = nn.Linear(size + size, size)
            self.concat_linear2 = nn.Linear(size + size, size)
        self.layer_id = layer_id

        if args.get("version", "v4") == "v4":
            from funasr.models.sense_voice.rwkv_v4 import RWKVLayer
            from funasr.models.sense_voice.rwkv_v4 import RWKV_TimeMix as RWKV_Tmix
        elif args.get("version", "v5") == "v5":
            from funasr.models.sense_voice.rwkv_v5 import RWKVLayer
            from funasr.models.sense_voice.rwkv_v5 import RWKV_Tmix_x052 as RWKV_Tmix
        else:
            from funasr.models.sense_voice.rwkv_v6 import RWKVLayer
            from funasr.models.sense_voice.rwkv_v6 import RWKV_Tmix_x060 as RWKV_Tmix
        # self.attn = RWKVLayer(args=args, layer_id=layer_id)
        self.self_attn = RWKV_Tmix(args, layer_id=layer_id)

        self.args = args
        self.ln0 = None
        if self.layer_id == 0 and not args.get("ln0", True):
            self.ln0 = LayerNorm(args.n_embd)
@@ -86,13 +107,25 @@
                layer_id = 0
                scale = ((1 + layer_id) / args.get("n_layer")) ** 0.7
                nn.init.constant_(self.ln0.weight, scale)
            

        # init
        if args.get("init_rwkv", True):
            print("init_rwkv")
            scale = ((1 + layer_id) / args.get("n_layer")) ** 0.7
            nn.init.constant_(self.norm1.weight, scale)
            nn.init.constant_(self.self_attn.ln2.weight, scale)
            # nn.init.constant_(self.self_attn.ln2.weight, scale)

        if args.get("init_rwkv", True):
            print("init_rwkv")
            nn.init.orthogonal_(self.self_attn.receptance.weight, gain=1)
            nn.init.orthogonal_(self.self_attn.key.weight, gain=0.1)
            nn.init.orthogonal_(self.self_attn.value.weight, gain=1)
            nn.init.orthogonal_(self.self_attn.gate.weight, gain=0.1)
            nn.init.zeros_(self.self_attn.output.weight)

        if args.get("datatype", "bf16") == "bf16":
            self.self_attn.to(torch.bfloat16)
            # self.norm1.to(torch.bfloat16)

    def forward(self, tgt, tgt_mask, memory, memory_mask, cache=None):
        """Compute decoded features.
@@ -115,28 +148,27 @@

        if self.layer_id == 0 and self.ln0 is not None:
            tgt = self.ln0(tgt)
            

        if self.args.get("datatype", "bf16") == "bf16":
            tgt = tgt.bfloat16()
        residual = tgt
        
        

        tgt = self.norm1(tgt)

        if cache is None:
    

            x = residual + self.dropout(self.self_attn(tgt, mask=tgt_mask))
        else:
            

            # tgt_q = tgt[:, -1:, :]
            # residual_q = residual[:, -1:, :]
            tgt_q_mask = None
            

            x = residual + self.dropout(self.self_attn(tgt, mask=tgt_q_mask))
            x = x[:, -1, :]



        if self.args.get("datatype", "bf16") == "bf16":
            x = x.to(torch.float32)
        # x = residual + self.dropout(self.self_attn(tgt_q, tgt, tgt, tgt_q_mask))
        

        residual = x
        x = self.norm2(x)
@@ -145,12 +177,10 @@
        x = self.norm3(x)
        x = residual + self.dropout(self.feed_forward(x))


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

        return x, tgt_mask, memory, memory_mask



class BaseTransformerDecoder(nn.Module, BatchScorerInterface):
@@ -176,15 +206,15 @@
    """

    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
@@ -217,11 +247,11 @@
        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.

@@ -249,20 +279,14 @@
        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:
@@ -272,11 +296,11 @@
        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.

@@ -296,9 +320,7 @@
            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:
@@ -313,13 +335,11 @@
    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.

@@ -343,8 +363,7 @@
        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
@@ -355,25 +374,26 @@
        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", "TransformerRWKVDecoder")
class TransformerRWKVDecoder(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,
            **kwargs,
        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,
        **kwargs,
    ):
        super().__init__(
            vocab_size=vocab_size,
@@ -385,19 +405,17 @@
            pos_enc_class=pos_enc_class,
            normalize_before=normalize_before,
        )
        from funasr.models.sense_voice.rwkv_v6 import RWKVLayer
        # from funasr.models.sense_voice.rwkv_v6 import RWKVLayer

        rwkv_cfg = kwargs.get("rwkv_cfg", {})
        args = OmegaConf.create(rwkv_cfg)
        # self.attn = RWKVLayer(args=args, layer_id=layer_id)

        attention_dim = encoder_output_size
        self.decoders = repeat(
            num_blocks,
            lambda lnum: DecoderLayer(
                attention_dim,
                RWKVLayer(args=args, layer_id=lnum),
                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,
@@ -406,18 +424,18 @@
                args=args,
            ),
        )
        

        # init
        if args.get("init_rwkv", True):
            print("init_rwkv")
            nn.init.uniform_(self.embed[0].weight, a=-1e-4, b=1e-4)

    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.

@@ -445,20 +463,14 @@
        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:
@@ -468,11 +480,11 @@
        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.

@@ -492,9 +504,7 @@
            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:
@@ -504,4 +514,4 @@
        if self.output_layer is not None:
            y = torch.log_softmax(self.output_layer(y), dim=-1)

        return y, new_cache
        return y, new_cache