alinashrestha/patentsbert-green-gold

PatentSBERTa Fine-Tuned with Gold Labels (Part D)

Task

Binary classification:

• —1 = Green patent
• —0 = Non-green patent

Model

This model fine-tunes AI-Growth-Lab/PatentSBERTa for binary classification (green vs non-green patents).

Training data:

• —train_silver
• —gold_100 (gold labels override silver for those 100 samples)

Training Settings

• —maxseqlength = 256
• —epochs = 1
• —learning_rate = 2e-5
• —batch_size = 16
• —GPU: NVIDIA L4

Results

Eval Silver

• —Accuracy: 0.8032
• —Precision: 0.8009
• —Recall: 0.8070
• —F1: 0.8039

Gold 100

• —Accuracy: 0.49
• —Precision: 0.0577
• —Recall: 0.60
• —F1: 0.1053

Dataset

Dataset available at: https://huggingface.co/datasets/alinashrestha/patents50kgreen.parquet

Notes

Model trained once as required in Part D.