/home/claude/source/sqnd-probe/BIP_Colab_Experiment.py:739:            return_tensors='pt'
/home/claude/source/sqnd-probe/BIP_Colab_Experiment.py:757:        'time_labels': torch.tensor([b['time_label'] for b in batch]),
/home/claude/source/sqnd-probe/BIP_Colab_Experiment.py:758:        'hohfeld_labels': torch.tensor([b['hohfeld_label'] for b in batch])
/home/claude/source/sqnd-probe/BIP_Colab_Experiment.py:952:    The bond embedding (z_bond) successfully captured moral structure
/home/claude/source/sqnd-probe/cell1_backbone_param.py:6:BACKBONE = "MiniLM"  #@param ["MiniLM", "LaBSE", "XLM-R-base", "XLM-R-large"]
/home/claude/source/sqnd-probe/cell1_backbone_param.py:8:#@markdown - **LaBSE**: Best cross-lingual alignment, 471M params (LaBSE)
/home/claude/source/sqnd-probe/cell1_backbone_param.py:19:    "LaBSE": {
/home/claude/source/sqnd-probe/cell1_backbone_param.py:20:        "model_name": "sentence-transformers/LaBSE",
/home/claude/source/sqnd-probe/cell1_v10.9.py:18:BACKBONE = "MiniLM"  # @param ["MiniLM", "LaBSE", "XLM-R-base", "XLM-R-large"]
/home/claude/source/sqnd-probe/cell1_v10.9.py:20:# @markdown - **LaBSE**: Best cross-lingual alignment, 471M params (recommended)
/home/claude/source/sqnd-probe/cell1_v10.9.py:43:    "LaBSE": {
/home/claude/source/sqnd-probe/cell1_v10.9.py:44:        "model_name": "sentence-transformers/LaBSE",
/home/claude/source/sqnd-probe/cell6_backbone_model.py:95:            # BERT-style models (XLM-R, LaBSE) have pooler_output
/home/claude/source/sqnd-probe/cell6_backbone_model.py:163:                            padding='max_length', return_tensors='pt')
/home/claude/source/sqnd-probe/cell6_backbone_model.py:182:        'bond_labels': torch.tensor([x['bond_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_backbone_model.py:183:        'language_labels': torch.tensor([x['language_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_backbone_model.py:184:        'period_labels': torch.tensor([x['period_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_backbone_model.py:185:        'hohfeld_labels': torch.tensor([x['hohfeld_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_backbone_model.py:186:        'context_labels': torch.tensor([x['context_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_backbone_model.py:187:        'sample_weights': torch.tensor([x['sample_weight'] for x in batch], dtype=torch.float),
/home/claude/source/sqnd-probe/cell6_temp.py:127:                            padding='max_length', return_tensors='pt')
/home/claude/source/sqnd-probe/cell6_temp.py:146:        'bond_labels': torch.tensor([x['bond_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_temp.py:147:        'language_labels': torch.tensor([x['language_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_temp.py:148:        'period_labels': torch.tensor([x['period_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_temp.py:149:        'hohfeld_labels': torch.tensor([x['hohfeld_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_temp.py:150:        'context_labels': torch.tensor([x['context_label'] for x in batch]),
/home/claude/source/sqnd-probe/cell6_temp.py:151:        'sample_weights': torch.tensor([x['sample_weight'] for x in batch], dtype=torch.float),
/home/claude/source/sqnd-probe/cell6_v10.9.py:172:    def get_bond_embedding(self, input_ids, attention_mask):
/home/claude/source/sqnd-probe/cell6_v10.9.py:173:        """Get z_bond embedding for geometric analysis."""
/home/claude/source/sqnd-probe/cell6_v10.9.py:231:            return_tensors="pt",
/home/claude/source/sqnd-probe/cell6_v10.9.py:253:        "bond_labels": torch.tensor([x["bond_label"] for x in batch]),
/home/claude/source/sqnd-probe/cell6_v10.9.py:254:        "language_labels": torch.tensor([x["language_label"] for x in batch]),
/home/claude/source/sqnd-probe/cell6_v10.9.py:255:        "period_labels": torch.tensor([x["period_label"] for x in batch]),
/home/claude/source/sqnd-probe/cell6_v10.9.py:256:        "hohfeld_labels": torch.tensor([x["hohfeld_label"] for x in batch]),
/home/claude/source/sqnd-probe/cell6_v10.9.py:257:        "context_labels": torch.tensor([x["context_label"] for x in batch]),
/home/claude/source/sqnd-probe/cell6_v10.9.py:258:        "sample_weights": torch.tensor([x["sample_weight"] for x in batch], dtype=torch.float),
/home/claude/source/sqnd-probe/cell7_v10.9.py:104:    # Clear any existing models/tensors from globals
/home/claude/source/sqnd-probe/cell7_v10.9.py:202:# @markdown Minimum embedding distance for role-swapped pairs
/home/claude/source/sqnd-probe/cell7_v10.9.py:460:            loss_role = torch.tensor(0.0, device=device)
/home/claude/source/sqnd-probe/cell7_v10.9.py:481:                        return_tensors="pt",
