# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Any
from typing import List
from typing import Tuple
import numpy as np
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from deepspeech.decoders.scorers.scorer_interface import BatchScorerInterface
from deepspeech.models.lm_interface import LMInterface
from deepspeech.modules.encoder import TransformerEncoder
from deepspeech.modules.mask import subsequent_mask
from deepspeech.utils.log import Log
logger = Log(__name__).getlog()
class TransformerLM(nn.Layer, LMInterface, BatchScorerInterface):
def __init__(self,
n_vocab: int,
pos_enc: str=None,
embed_unit: int=128,
att_unit: int=256,
head: int=2,
unit: int=1024,
layer: int=4,
dropout_rate: float=0.5,
emb_dropout_rate: float=0.0,
att_dropout_rate: float=0.0,
tie_weights: bool=False,
**kwargs):
nn.Layer.__init__(self)
if pos_enc == "sinusoidal":
pos_enc_layer_type = "abs_pos"
elif pos_enc is None:
pos_enc_layer_type = "no_pos"
else:
raise ValueError(f"unknown pos-enc option: {pos_enc}")
self.embed = nn.Embedding(n_vocab, embed_unit)
if emb_dropout_rate == 0.0:
self.embed_drop = None
else:
self.embed_drop = nn.Dropout(emb_dropout_rate)
self.encoder = TransformerEncoder(
input_size=embed_unit,
output_size=att_unit,
attention_heads=head,
linear_units=unit,
num_blocks=layer,
dropout_rate=dropout_rate,
attention_dropout_rate=att_dropout_rate,
input_layer="linear",
pos_enc_layer_type=pos_enc_layer_type,
concat_after=False,
static_chunk_size=1,
use_dynamic_chunk=False,
use_dynamic_left_chunk=False)
self.decoder = nn.Linear(att_unit, n_vocab)
logger.info("Tie weights set to {}".format(tie_weights))
logger.info("Dropout set to {}".format(dropout_rate))
logger.info("Emb Dropout set to {}".format(emb_dropout_rate))
logger.info("Att Dropout set to {}".format(att_dropout_rate))
if tie_weights:
assert (
att_unit == embed_unit
), "Tie Weights: True need embedding and final dimensions to match"
self.decoder.weight = self.embed.weight
def _target_mask(self, ys_in_pad):
ys_mask = ys_in_pad != 0
m = subsequent_mask(ys_mask.size(-1)).unsqueeze(0)
return ys_mask.unsqueeze(-2) & m
def forward(self, x: paddle.Tensor, t: paddle.Tensor
) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
"""Compute LM loss value from buffer sequences.
Args:
x (paddle.Tensor): Input ids. (batch, len)
t (paddle.Tensor): Target ids. (batch, len)
Returns:
tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]: Tuple of
loss to backward (scalar),
negative log-likelihood of t: -log p(t) (scalar) and
the number of elements in x (scalar)
Notes:
The last two return values are used
in perplexity: p(t)^{-n} = exp(-log p(t) / n)
"""
batch_size = x.size(0)
xm = x != 0
xlen = xm.sum(axis=1)
if self.embed_drop is not None:
emb = self.embed_drop(self.embed(x))
else:
emb = self.embed(x)
h, _ = self.encoder(emb, xlen)
y = self.decoder(h)
loss = F.cross_entropy(
y.view(-1, y.shape[-1]), t.view(-1), reduction="none")
mask = xm.to(loss.dtype)
logp = loss * mask.view(-1)
nll = logp.view(batch_size, -1).sum(-1)
nll_count = mask.sum(-1)
logp = logp.sum()
count = mask.sum()
return logp / count, logp, count, nll, nll_count
# beam search API (see ScorerInterface)
def score(self, y: paddle.Tensor, state: Any,
x: paddle.Tensor) -> Tuple[paddle.Tensor, Any]:
"""Score new token.
Args:
y (paddle.Tensor): 1D paddle.int64 prefix tokens.
state: Scorer state for prefix tokens
x (paddle.Tensor): encoder feature that generates ys.
Returns:
tuple[paddle.Tensor, Any]: Tuple of
paddle.float32 scores for next token (n_vocab)
and next state for ys
"""
y = y.unsqueeze(0)
if self.embed_drop is not None:
emb = self.embed_drop(self.embed(y))
else:
emb = self.embed(y)
h, _, cache = self.encoder.forward_one_step(
emb, self._target_mask(y), cache=state)
h = self.decoder(h[:, -1])
logp = F.log_softmax(h).squeeze(0)
return logp, cache
# batch beam search API (see BatchScorerInterface)
def batch_score(self,
ys: paddle.Tensor,
states: List[Any],
xs: paddle.Tensor) -> Tuple[paddle.Tensor, List[Any]]:
"""Score new token batch (required).
Args:
ys (paddle.Tensor): paddle.int64 prefix tokens (n_batch, ylen).
states (List[Any]): Scorer states for prefix tokens.
xs (paddle.Tensor):
The encoder feature that generates ys (n_batch, xlen, n_feat).
Returns:
tuple[paddle.Tensor, List[Any]]: Tuple of
batchfied scores for next token with shape of `(n_batch, n_vocab)`
and next state list for ys.
"""
# merge states
n_batch = len(ys)
n_layers = len(self.encoder.encoders)
if states[0] is None:
batch_state = None
else:
# transpose state of [batch, layer] into [layer, batch]
batch_state = [
paddle.stack([states[b][i] for b in range(n_batch)])
for i in range(n_layers)
]
if self.embed_drop is not None:
emb = self.embed_drop(self.embed(ys))
else:
emb = self.embed(ys)
# batch decoding
h, _, states = self.encoder.forward_one_step(
emb, self._target_mask(ys), cache=batch_state)
h = self.decoder(h[:, -1])
logp = F.log_softmax(h)
# transpose state of [layer, batch] into [batch, layer]
state_list = [[states[i][b] for i in range(n_layers)]
for b in range(n_batch)]
return logp, state_list
if __name__ == "__main__":
tlm = TransformerLM(
n_vocab=5002,
pos_enc=None,
embed_unit=128,
att_unit=512,
head=8,
unit=2048,
layer=16,
dropout_rate=0.5, )
# n_vocab: int,
# pos_enc: str=None,
# embed_unit: int=128,
# att_unit: int=256,
# head: int=2,
# unit: int=1024,
# layer: int=4,
# dropout_rate: float=0.5,
# emb_dropout_rate: float = 0.0,
# att_dropout_rate: float = 0.0,
# tie_weights: bool = False,):
paddle.set_device("cpu")
model_dict = paddle.load("transformerLM.pdparams")
tlm.set_state_dict(model_dict)
tlm.eval()
#Test the score
input2 = np.array([5])
input2 = paddle.to_tensor(input2)
state = None
output, state = tlm.score(input2, state, None)
input3 = np.array([5, 10])
input3 = paddle.to_tensor(input3)
output, state = tlm.score(input3, state, None)
input4 = np.array([5, 10, 0])
input4 = paddle.to_tensor(input4)
output, state = tlm.score(input4, state, None)
print("output", output)
"""
#Test the batch score
batch_size = 2
inp2 = np.array([[5], [10]])
inp2 = paddle.to_tensor(inp2)
output, states = tlm.batch_score(
inp2, [(None,None,0)] * batch_size)
inp3 = np.array([[100], [30]])
inp3 = paddle.to_tensor(inp3)
output, states = tlm.batch_score(
inp3, states)
print("output", output)
#print("cache", cache)
#np.save("output_pd.npy", output)
"""