Python utils.make_positions方法代碼示例

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(self, input, incremental_state=None):
        """Input is expected to be of size [bsz x seqlen]."""
        # recompute/expand embeddings if needed
        bsz, seq_len = input.size()
        max_pos = self.padding_idx + 1 + seq_len
        if max_pos > self.weights.size(0):
            self.weights = SinusoidalPositionalEmbedding.get_embedding(
                max_pos,
                self.embedding_dim,
                self.padding_idx,
            ).type_as(self.weights)
        self.weights = self.weights.type_as(self._float_tensor)

        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            return self.weights[self.padding_idx + seq_len, :].expand(bsz, 1, -1)

        positions = utils.make_positions(input.data, self.padding_idx, self.left_pad)
        return self.weights.index_select(0, positions.view(-1)).view(bsz, seq_len, -1) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(self, input, incremental_state=None):
        """Input is expected to be of size [bsz x seqlen]."""
        # recompute/expand embeddings if needed
        bsz, seq_len = input.size()
        max_pos = self.padding_idx + 1 + seq_len
        if self.weights is None or max_pos > self.weights.size(0):
            self.weights = SinusoidalPositionalEmbedding.get_embedding(
                max_pos,
                self.embedding_dim,
                self.padding_idx,
            )
        self.weights = self.weights.type_as(self._float_tensor)

        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            return self.weights[self.padding_idx + seq_len, :].expand(bsz, 1, -1)

        positions = utils.make_positions(input.data, self.padding_idx, self.left_pad)
        return self.weights.index_select(0, positions.view(-1)).view(bsz, seq_len, -1).detach() 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def test_make_positions(self):
        pad = 1
        left_pad_input = torch.LongTensor([
            [9, 9, 9, 9, 9],
            [1, 9, 9, 9, 9],
            [1, 1, 1, 9, 9],
        ])
        left_pad_output = torch.LongTensor([
            [2, 3, 4, 5, 6],
            [1, 2, 3, 4, 5],
            [1, 1, 1, 2, 3],
        ])
        right_pad_input = torch.LongTensor([
            [9, 9, 9, 9, 9],
            [9, 9, 9, 9, 1],
            [9, 9, 1, 1, 1],
        ])
        right_pad_output = torch.LongTensor([
            [2, 3, 4, 5, 6],
            [2, 3, 4, 5, 1],
            [2, 3, 1, 1, 1],
        ])

        self.assertAlmostEqual(
            left_pad_output,
            utils.make_positions(left_pad_input, pad, left_pad=True),
        )
        self.assertAlmostEqual(
            right_pad_output,
            utils.make_positions(right_pad_input, pad, left_pad=False),
        ) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(self, input, incremental_state=None):
        """Input is expected to be of size [bsz x seqlen]."""
        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            positions = input.data.new(1, 1).fill_(self.padding_idx + input.size(1))
        else:
            positions = utils.make_positions(input.data, self.padding_idx, self.left_pad)
        return super().forward(positions) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def test_make_positions(self):
        pad = 1
        left_pad_input = torch.LongTensor([
            [9, 9, 9, 9, 9],
            [1, 9, 9, 9, 9],
            [1, 1, 1, 9, 9],
        ])
        left_pad_output = torch.LongTensor([
            [2, 3, 4, 5, 6],
            [1, 2, 3, 4, 5],
            [1, 1, 1, 2, 3],
        ])
        right_pad_input = torch.LongTensor([
            [9, 9, 9, 9, 9],
            [9, 9, 9, 9, 1],
            [9, 9, 1, 1, 1],
        ])
        right_pad_output = torch.LongTensor([
            [2, 3, 4, 5, 6],
            [2, 3, 4, 5, 1],
            [2, 3, 1, 1, 1],
        ])

        self.assertAlmostEqual(
            left_pad_output,
            utils.make_positions(left_pad_input, pad),
        )
        self.assertAlmostEqual(
            right_pad_output,
            utils.make_positions(right_pad_input, pad),
        ) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(
        self,
        input: Tensor,
        incremental_state: Optional[Dict[str, Dict[str, Optional[Tensor]]]] = None,
        positions: Optional[Tensor] = None,
    ):
        """Input is expected to be of size [bsz x seqlen]."""
        assert (positions is None) or (
            self.padding_idx is None
        ), "If positions is pre-computed then padding_idx should not be set."

        if positions is None:
            if incremental_state is not None:
                # positions is the same for every token when decoding a single step
                # Without the int() cast, it doesn't work in some cases when exporting to ONNX
                positions = torch.zeros(
                    (1, 1), device=input.device, dtype=input.dtype
                ).fill_(int(self.padding_idx + input.size(1)))
            else:
                positions = utils.make_positions(
                    input, self.padding_idx, onnx_trace=self.onnx_trace
                )
        return F.embedding(
            positions,
            self.weight,
            self.padding_idx,
            self.max_norm,
            self.norm_type,
            self.scale_grad_by_freq,
            self.sparse,
        ) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(self, input, incremental_state=None):
        """Input is expected to be of size [bsz x seqlen]."""
        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            positions = input.data.new(1, 1).fill_(self.padding_idx + input.size(1))
        else:
            positions = utils.make_positions(input.data, self.padding_idx, self.left_pad, self.onnx_trace)
        return super().forward(positions) 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(self, input, incremental_state=None, timestep=None):
        """Input is expected to be of size [bsz x seqlen]."""
        bsz, seq_len = torch.onnx.operators.shape_as_tensor(input)
        max_pos = self.padding_idx + 1 + seq_len
        if self.weights is None or max_pos > self.weights.size(0):
            # recompute/expand embeddings if needed
            self.weights = SinusoidalPositionalEmbedding.get_embedding(
                max_pos,
                self.embedding_dim,
                self.padding_idx,
            )
        self.weights = self.weights.type_as(self._float_tensor)

        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            pos = (timestep.int() + 1).long() if timestep is not None else seq_len
            if self.onnx_trace:
                return self.weights[self.padding_idx + pos, :].unsqueeze(1).repeat(bsz, 1, 1)
            return self.weights[self.padding_idx + pos, :].expand(bsz, 1, -1)

        positions = utils.make_positions(input, self.padding_idx, self.left_pad, self.onnx_trace)
        if self.onnx_trace:
            flat_embeddings = self.weights.detach().index_select(0, positions.view(-1))
            embedding_shape = torch.cat((bsz.view(1), seq_len.view(1), torch.LongTensor([-1])))
            embeddings = torch.onnx.operators.reshape_from_tensor_shape(flat_embeddings, embedding_shape)
            return embeddings
        return self.weights.index_select(0, positions.view(-1)).view(bsz, seq_len, -1).detach() 

# 需要導入模塊: from fairseq import utils [as 別名]
# 或者: from fairseq.utils import make_positions [as 別名]
def forward(
        self,
        input,
        incremental_state: Optional[Any] = None,
        timestep: Optional[Tensor] = None,
        positions: Optional[Any] = None,
    ):
        """Input is expected to be of size [bsz x seqlen]."""
        bspair = torch.onnx.operators.shape_as_tensor(input)
        bsz, seq_len = bspair[0], bspair[1]
        max_pos = self.padding_idx + 1 + seq_len
        if self.weights is None or max_pos > self.weights.size(0):
            # recompute/expand embeddings if needed
            self.weights = SinusoidalPositionalEmbedding.get_embedding(
                max_pos, self.embedding_dim, self.padding_idx
            )
        self.weights = self.weights.to(self._float_tensor)

        if incremental_state is not None:
            # positions is the same for every token when decoding a single step
            pos = timestep.view(-1)[0] + 1 if timestep is not None else seq_len
            if self.onnx_trace:
                return (
                    self.weights.index_select(index=self.padding_idx + pos, dim=0)
                    .unsqueeze(1)
                    .repeat(bsz, 1, 1)
                )
            return self.weights[self.padding_idx + pos, :].expand(bsz, 1, -1)

        positions = utils.make_positions(
            input, self.padding_idx, onnx_trace=self.onnx_trace
        )
        if self.onnx_trace:
            flat_embeddings = self.weights.detach().index_select(0, positions.view(-1))
            embedding_shape = torch.cat(
                (bsz.view(1), seq_len.view(1), torch.tensor([-1], dtype=torch.long))
            )
            embeddings = torch.onnx.operators.reshape_from_tensor_shape(
                flat_embeddings, embedding_shape
            )
            return embeddings
        return (
            self.weights.index_select(0, positions.view(-1))
            .view(bsz, seq_len, -1)
            .detach()
        )