File size: 7,144 Bytes
6fc683c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 |
# Copyright (c) Facebook, Inc. and its affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
import math
from dataclasses import dataclass, field
from typing import List, Optional
import torch
import torch.nn.functional as F
from fairseq import metrics, utils
from fairseq.criterions import FairseqCriterion, register_criterion
from fairseq.dataclass import FairseqDataclass
from fairseq.logging.meters import safe_round
@dataclass
class Wav2VecCriterionConfig(FairseqDataclass):
infonce: bool = field(
default=False,
metadata={
"help": "if set, uses cross entropy instead of binary cross entropy (i.e. InfoNCE loss)"
},
)
loss_weights: Optional[List[float]] = field(
default=None,
metadata={"help": "weights for additional loss terms (not first one)"},
)
log_keys: List[str] = field(
default_factory=lambda: [],
metadata={"help": "output keys to log"},
)
@register_criterion("wav2vec", dataclass=Wav2VecCriterionConfig)
class Wav2vecCriterion(FairseqCriterion):
def __init__(self, task, infonce=False, loss_weights=None, log_keys=None):
super().__init__(task)
self.infonce = infonce
self.loss_weights = loss_weights
self.log_keys = [] if log_keys is None else log_keys
def forward(self, model, sample, reduce=True):
"""Compute the loss for the given sample.
Returns a tuple with three elements:
1) the loss
2) the sample size, which is used as the denominator for the gradient
3) logging outputs to display while training
"""
net_output = model(**sample["net_input"])
logits = model.get_logits(net_output).float()
target = model.get_targets(sample, net_output)
weights = None
if hasattr(model, "get_target_weights") and not self.infonce:
weights = model.get_target_weights(target, net_output)
if torch.is_tensor(weights):
weights = weights.float()
losses = []
if self.infonce:
loss = F.cross_entropy(
logits,
target,
reduction="sum" if reduce else "none",
)
else:
loss = F.binary_cross_entropy_with_logits(
logits,
target.float(),
weights,
reduction="sum" if reduce else "none",
)
sample_size = target.numel() if self.infonce else target.long().sum().item()
losses.append(loss.detach().clone())
if self.loss_weights is not None:
assert hasattr(model, "get_extra_losses")
extra_losses = model.get_extra_losses(net_output)
if torch.is_tensor(extra_losses):
extra_losses = [extra_losses]
if len(self.loss_weights) == 1 and len(extra_losses) != 1:
self.loss_weights = [self.loss_weights[0]] * len(extra_losses)
assert len(extra_losses) == len(
self.loss_weights
), f"{len(extra_losses)}, {len(self.loss_weights)}"
for p, coef in zip(extra_losses, self.loss_weights):
if coef != 0 and p is not None:
p = coef * p.float() * sample_size
loss += p
losses.append(p)
logging_output = {
"loss": loss.item() if reduce else loss,
"ntokens": sample_size,
"nsentences": sample["id"].numel(),
"sample_size": sample_size,
}
for lk in self.log_keys:
# Only store "logits" and "target" for computing MAP and MAUC
# during validation
if lk == "logits":
if not self.training:
logging_output["logits"] = logits.cpu().numpy()
elif lk == "target":
if not self.training:
logging_output["target"] = target.cpu().numpy()
elif lk in net_output:
logging_output[lk] = float(net_output[lk])
if len(losses) > 1:
for i, l in enumerate(losses):
logging_output[f"loss_{i}"] = l.item()
if self.infonce:
with torch.no_grad():
if logits.numel() == 0:
corr = 0
count = 0
else:
assert logits.dim() > 1, logits.shape
max = logits.argmax(-1) == 0
min = logits.argmin(-1) == 0
both = max & min
corr = max.long().sum().item() - both.long().sum().item()
count = max.numel()
logging_output["correct"] = corr
logging_output["count"] = count
return loss, sample_size, logging_output
@staticmethod
def reduce_metrics(logging_outputs) -> None:
"""Aggregate logging outputs from data parallel training."""
loss_sum = utils.item(sum(log.get("loss", 0) for log in logging_outputs))
ntokens = utils.item(sum(log.get("ntokens", 0) for log in logging_outputs))
nsentences = utils.item(
sum(log.get("nsentences", 0) for log in logging_outputs)
)
sample_size = utils.item(
sum(log.get("sample_size", 0) for log in logging_outputs)
)
metrics.log_scalar(
"loss", loss_sum / (sample_size or 1) / math.log(2), sample_size, round=3
)
metrics.log_scalar("ntokens", ntokens)
metrics.log_scalar("nsentences", nsentences)
correct = sum(log.get("correct", 0) for log in logging_outputs)
metrics.log_scalar("_correct", correct)
total = sum(log.get("count", 0) for log in logging_outputs)
metrics.log_scalar("_total", total)
if total > 0:
metrics.log_derived(
"accuracy",
lambda meters: safe_round(
meters["_correct"].sum / meters["_total"].sum, 5
)
if meters["_total"].sum > 0
else float("nan"),
)
builtin_keys = {
"loss",
"ntokens",
"nsentences",
"sample_size",
"correct",
"count",
}
for k in logging_outputs[0]:
if k not in builtin_keys:
val = sum(log.get(k, 0) for log in logging_outputs)
if k.startswith("loss"):
metrics.log_scalar(
k, val / (sample_size or 1) / math.log(2), sample_size, round=3
)
else:
metrics.log_scalar(k, val / len(logging_outputs), round=3)
@staticmethod
def logging_outputs_can_be_summed() -> bool:
"""
Whether the logging outputs returned by `forward` can be summed
across workers prior to calling `reduce_metrics`. Setting this
to True will improves distributed training speed.
"""
return False
|