File size: 8,179 Bytes
1de51cd | 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 | from typing import Any, Tuple, Dict, Sequence, Optional
import torch
import torch.nn.functional as F
from torch import nn
import math
IGNORE_LABEL_ID = -100
def s(x, epsilon=1e-30):
return torch.where(
x<0,
1/(1-x+ epsilon),
x + 1
)
def log_stablemax(x, dim=-1):
s_x = s(x)
return torch.log(s_x/torch.sum(s_x, dim=dim, keepdim=True))
def stablemax_cross_entropy(logits, labels, ignore_index: int = -100, valid_mask=None):
logprobs = log_stablemax(logits.to(torch.float64), dim=-1)
if valid_mask is None:
valid_mask = (labels != ignore_index)
transformed_labels = torch.where(valid_mask, labels, 0)
prediction_logprobs = torch.gather(logprobs, index=transformed_labels.to(torch.long).unsqueeze(-1), dim=-1).squeeze(-1)
return -torch.where(valid_mask, prediction_logprobs, 0)
def softmax_cross_entropy(logits, labels, ignore_index: int = -100):
# Cast logits to f32
# Flatten logits
return F.cross_entropy(logits.to(torch.float32).view(-1, logits.shape[-1]), labels.to(torch.long).view(-1), ignore_index=ignore_index, reduction="none").view(labels.shape)
class ACTLossHead(nn.Module):
def __init__(self, model: nn.Module, loss_type: str, alpha: float = 0.5):
super().__init__()
self.model = model
self.loss_fn = globals()[loss_type]
def initial_carry(self, *args, **kwargs):
return self.model.initial_carry(*args, **kwargs) # type: ignore
@staticmethod
def comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts):
q_halt_loss = F.binary_cross_entropy_with_logits(outputs["q_halt_logits"], seq_is_correct.to(outputs["q_halt_logits"].dtype), reduction="sum")
return q_halt_loss
def forward(
self,
return_keys: Sequence[str],
# Model args
**model_kwargs,
) -> Tuple[Any, torch.Tensor, Dict[str, torch.Tensor], Optional[Dict[str, torch.Tensor]], torch.Tensor]:
# Model logits
# B x SeqLen x D
new_carry, outputs = self.model(**model_kwargs)
labels = new_carry.current_data["labels"]
with torch.no_grad():
# Preds
outputs["preds"] = torch.argmax(outputs["logits"], dim=-1)
# Correctness
mask = (labels != IGNORE_LABEL_ID)
loss_counts = mask.sum(-1)
loss_divisor = loss_counts.clamp_min(1).unsqueeze(-1) # Avoid NaNs in division
is_correct = mask & (torch.argmax(outputs["logits"], dim=-1) == labels)
seq_is_correct = is_correct.sum(-1) == loss_counts
# Metrics (halted)
valid_metrics = new_carry.halted & (loss_counts > 0)
metrics = {
"count": valid_metrics.sum(),
"accuracy": torch.where(valid_metrics, (is_correct.to(torch.float32) / loss_divisor).sum(-1), 0).sum(),
"exact_accuracy": (valid_metrics & seq_is_correct).sum(),
"q_halt_accuracy": (valid_metrics & ((outputs["q_halt_logits"] >= 0) == seq_is_correct)).sum(),
"steps": torch.where(valid_metrics, new_carry.steps, 0).sum(),
}
# Losses
lm_loss = (self.loss_fn(outputs["logits"], labels, ignore_index=IGNORE_LABEL_ID, valid_mask=mask) / loss_divisor).sum()
q_halt_loss = self.comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts)
metrics.update({
"lm_loss": lm_loss.detach(),
"q_halt_loss": q_halt_loss.detach(),
})
# Q continue (bootstrapping target loss); Alexia: This fits Q-learning, but seems totally unecessary
q_continue_loss = 0
if "target_q_continue" in outputs:
q_continue_loss = F.binary_cross_entropy_with_logits(outputs["q_continue_logits"], outputs["target_q_continue"], reduction="sum")
metrics["q_continue_loss"] = q_continue_loss.detach()
# Filter outputs for return
detached_outputs = {k: outputs[k].detach() for k in return_keys if k in outputs}
return new_carry, lm_loss + 0.5 * (q_halt_loss + q_continue_loss), metrics, detached_outputs, new_carry.halted.all()
class ACTLossHead_binorminal(ACTLossHead):
def __init__(self, model: nn.Module, loss_type: str, alpha: float = 0.5):
super().__init__(model, loss_type)
@staticmethod
def comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts):
t = (is_correct.sum(-1) / loss_counts.clamp_min(1)).to(outputs["q_halt_logits"].dtype) # [B] in [0,1]
# Force t=1 for sequences with no valid tokens
t = torch.where(loss_counts == 0, torch.ones_like(t), t)
q_halt_loss = F.binary_cross_entropy_with_logits(
outputs["q_halt_logits"], t, reduction="sum"
)
return q_halt_loss
class ACTLossHead_combine(ACTLossHead):
def __init__(self, model: nn.Module, loss_type: str, alpha: float = 0.5):
super().__init__(model, loss_type)
self.alpha = alpha
@staticmethod
def comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts):
q_halt_bce_loss = ACTLossHead.comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts)
q_halt_binorm_loss = ACTLossHead_binorminal.comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts)
q_halt_loss = self.alpha * q_halt_bce_loss + (1 - self.alpha) * q_halt_binorm_loss
return q_halt_loss
class ACTLossHead_binormonly(ACTLossHead_binorminal):
def __init__(self, model: nn.Module, loss_type: str):
super().__init__(model, loss_type)
def forward(
self,
return_keys: Sequence[str],
# Model args
**model_kwargs,
) -> Tuple[Any, torch.Tensor, Dict[str, torch.Tensor], Optional[Dict[str, torch.Tensor]], torch.Tensor]:
# Model logits
# B x SeqLen x D
new_carry, outputs = self.model(**model_kwargs)
labels = new_carry.current_data["labels"]
with torch.no_grad():
# Preds
outputs["preds"] = torch.argmax(outputs["logits"], dim=-1)
# Correctness
mask = (labels != IGNORE_LABEL_ID)
loss_counts = mask.sum(-1)
loss_divisor = loss_counts.clamp_min(1).unsqueeze(-1) # Avoid NaNs in division
is_correct = mask & (torch.argmax(outputs["logits"], dim=-1) == labels)
seq_is_correct = is_correct.sum(-1) == loss_counts
# Metrics (halted)
valid_metrics = new_carry.halted & (loss_counts > 0)
metrics = {
"count": valid_metrics.sum(),
"accuracy": torch.where(valid_metrics, (is_correct.to(torch.float32) / loss_divisor).sum(-1), 0).sum(),
"exact_accuracy": (valid_metrics & seq_is_correct).sum(),
"q_halt_accuracy": (valid_metrics & ((outputs["q_halt_logits"] >= 0) == seq_is_correct)).sum(),
"steps": torch.where(valid_metrics, new_carry.steps, 0).sum(),
}
# Losses
lm_loss = (self.loss_fn(outputs["logits"], labels, ignore_index=IGNORE_LABEL_ID, valid_mask=mask) / loss_divisor).sum()
q_halt_loss = self.comput_q_halt_loss(self, outputs, is_correct, seq_is_correct, loss_counts)
metrics.update({
"lm_loss": lm_loss.detach(),
"q_halt_loss": q_halt_loss.detach(),
})
# Q continue (bootstrapping target loss); Alexia: This fits Q-learning, but seems totally unecessary
q_continue_loss = 0
if "target_q_continue" in outputs:
q_continue_loss = F.binary_cross_entropy_with_logits(outputs["q_continue_logits"], outputs["target_q_continue"], reduction="sum")
metrics["q_continue_loss"] = q_continue_loss.detach()
# Filter outputs for return
detached_outputs = {k: outputs[k].detach() for k in return_keys if k in outputs}
return new_carry, (q_halt_loss + q_continue_loss), metrics, detached_outputs, new_carry.halted.all()
|