Files
composable-evidence-states-…/losses/alignment_loss.py
T
2026-10-01 22:16:29 +08:00

37 lines
1.3 KiB
Python

from __future__ import annotations
import torch
import torch.nn.functional as F
def supervised_alignment_loss(
states: torch.Tensor,
targets: list[torch.Tensor],
assignments: list[tuple[torch.Tensor, torch.Tensor]],
temperature: float = 0.1,
) -> torch.Tensor:
vectors = []
labels = []
for batch_index, (slot_index, fact_index) in enumerate(assignments):
if len(slot_index) == 0:
continue
vectors.append(states[batch_index, slot_index])
labels.append(targets[batch_index].to(states.device)[fact_index])
if not vectors:
return states.sum() * 0.0
vectors_tensor = F.normalize(torch.cat(vectors), dim=-1)
labels_tensor = torch.cat(labels)
if len(vectors_tensor) < 2:
return vectors_tensor.sum() * 0.0
same = (labels_tensor[:, None, :] == labels_tensor[None, :, :]).all(-1)
same.fill_diagonal_(False)
logits = vectors_tensor @ vectors_tensor.T / temperature
logits.fill_diagonal_(-torch.inf)
log_prob = logits - torch.logsumexp(logits, dim=1, keepdim=True)
positive_count = same.sum(1)
usable = positive_count > 0
if not usable.any():
return vectors_tensor.sum() * 0.0
return -((log_prob.masked_fill(~same, 0.0).sum(1) / positive_count.clamp_min(1))[usable]).mean()