Peter512/patentsbert-green-a3

PatentSBERTa Green Patent Classifier — Assignment 3

Binary classifier for green patent detection (Y02 CPC codes). Fine-tuned from AI-Growth-Lab/PatentSBERTa using a 3-agent CrewAI debate system (Advocate / Skeptic / Judge).

Training

• Base model: AI-Growth-Lab/PatentSBERTa (MPNet-based)
• Task: Binary classification — is_green (Y02 CPC codes)
• Training data: 35,000 silver labels + 100 gold labels (CrewAI MAS)
• MAS process: 3-agent debate — Advocate argues green, Skeptic challenges, Judge produces {"label": 0/1, "confidence": "low/medium/high", "rationale": "..."}. 100% agent agreement (0 human overrides — no low-confidence outputs).
• Fine-tuning: 1 epoch, lr=2e-5, max_length=256, batch_size=16, fp16

Evaluation (eval_silver, 5,000 claims)

Metric	Value
F1	0.8115
Precision	0.8224
Recall	0.8010
Accuracy	0.8142

Assignment 2 baseline: F1=0.8099 | Original baseline: F1=0.7696

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("AI-Growth-Lab/PatentSBERTa", use_fast=False)
model = AutoModelForSequenceClassification.from_pretrained("Peter512/patentsbert-green-a3")
model.eval()

text = "A photovoltaic cell comprising a perovskite absorber layer..."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)
with torch.no_grad():
    logits = model(**inputs).logits
label = logits.argmax().item()  # 0=not_green, 1=green