yeseul0-0/docred-stage2-atlop

DocRED Stage2 Model (ATLOP + DREEAM)

Base Model

• bert-base-uncased (Apache 2.0)

Dataset

• DocRED (CC BY-SA 4.0)

Training

• Stage 2: distant pre-training + annotated fine-tuning
• pooling: logsumexp
• classifier: ATLOP
• threshold: adaptive
• Encoder: bert-base-uncased
• Max sequence length: 512
• Batch size: 4
• Optimizer: AdamW
• Encoder LR: 2e-5
• Classifier LR: 1e-4
• Warmup ratio: 0.06

Stage Enhancements

• Entity pooling: LogSumExp (ATLOP)
• Relation classifier: ATLOP (bilinear-style)
• Thresholding: Adaptive threshold
• Evidence supervision: Enabled (DREEAM)

License

This model is licensed under CC BY-SA 4.0 due to the use of the DocRED dataset.

• Base model: bert-base-uncased (Apache 2.0)
• Dataset: DocRED (CC BY-SA 4.0)