# Copyright 2022 Cerebras Systems.
#
# 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.
# Based on https://github.com/huggingface/trl/blob/main/trl/trainer/dpo_trainer.py
import torch
import torch.nn as nn
import torch.nn.functional as F
from cerebras.modelzoo.common.registry import registry
[docs]@registry.register_loss("DPOLoss")
class DPOLoss(nn.Module):
"""
DPO Loss
:param beta: Temperature parameter for the DPO loss, typically something in the range of 0.1 to 0.5.
We ignore the reference model as beta -> 0.
:param reference_free: If True, we ignore the _provided_ reference model and implicitly use a
reference model that assigns equal probability to all responses.
"""
def __init__(
self,
beta=0.1,
loss_type="sigmoid",
reference_free=False,
):
super(DPOLoss, self).__init__()
self.beta = beta
self.loss_type = loss_type
self.reference_free = reference_free
def forward(
self,
policy_chosen_logps,
policy_rejected_logps,
reference_chosen_logps,
reference_rejected_logps,
):
pi_logratios = policy_chosen_logps - policy_rejected_logps
if self.reference_free:
ref_logratios = 0
else:
ref_logratios = reference_chosen_logps - reference_rejected_logps
logits = pi_logratios - ref_logratios
if self.loss_type == "sigmoid":
losses = -F.logsigmoid(self.beta * logits)
elif self.loss_type == "hinge":
losses = torch.relu(1 - self.beta * logits)
elif self.loss_type == "ipo":
losses = torch.pow(logits - 1 / (2 * self.beta), 2.0)
else:
raise ValueError(
f"Unknown loss type: {self.loss_type}. Should be one of ['sigmoid', 'hinge', 'ipo']"
)
return losses.mean()