---
language:
- en
license: apache-2.0
base_model: mpi-inno-comp/paecter
library_name: transformers
pipeline_tag: text-classification
tags:
- patents
- plant-patents
- plant-breeding
- plant-varieties
- text-classification
- paecter
- reproducibility
widget:
- text: |-
    Drought tolerant maize plant

    A maize plant with improved drought tolerance is provided, together with seeds and methods for producing the plant.
- text: |-
    Rose plant named Sunny Belle

    A new and distinct cultivar of Rosa plant is described, characterized by compact growth and yellow flowers.
---

# PaECTER Variety-vs-Technology Patent-Family Classifier

This is the second-stage production classifier for the manuscript
*Identifying Plant-Related Patents: Corpus Construction and Global Patterns*.
It fine-tunes `mpi-inno-comp/paecter` to classify plant-related patent families
as variety-focused or technology-focused using English title and abstract text.

This release is a companion checkpoint for reproducible research on
variety-focused and technology-focused plant patent classification. The
Apache-2.0 license metadata follows the PaECTER base-model release.

## Model Details

- Base model: [`mpi-inno-comp/paecter`](https://huggingface.co/mpi-inno-comp/paecter)
- Architecture: `BertForSequenceClassification`
- Task: binary text classification
- Input text: English title, blank line, English abstract
- Maximum length: 512 tokens
- Labels:
  - `variety`: variety-focused plant patent family
  - `technology`: technology-focused plant patent family
- Recommended decision rule: classify as `technology` when
  `p_technology >= 0.000399`

This model is intended to be applied only after the first-stage plant-related
detector has selected plant-related families.

This model is a supervised fine-tune of PaECTER. Please cite or link the
original PaECTER model and paper when reusing this checkpoint.

## Training Data

The model was trained on the plant-positive subset of the repository's internal
600-family labeled DOCDB sample. The subtype split contains 142 train, 30
validation, and 31 test families. The row-level labeling queue, patent text,
split manifest, and prediction files are not part of this model release.

## Evaluation

Metrics below use the validation-F1-selected threshold
`p_technology >= 0.000399`, with `technology` as the positive class.

| Split | PR-AUC | Precision | Recall | F1 | Accuracy |
|---|---:|---:|---:|---:|---:|
| Validation | 0.9740 | 0.9500 | 1.0000 | 0.9744 | 0.9667 |
| Test | 1.0000 | 1.0000 | 1.0000 | 1.0000 | 1.0000 |

The held-out test split is small (`n = 31`), so differences between subtype
models should be interpreted cautiously.

## Intended Use

Use this model to separate already plant-related patent families into
variety-focused and technology-focused subtypes. It is useful for corpus
summaries, downstream descriptive analysis, and manual review workflows where
the input family has already passed the plant-related inclusion stage.

## Limitations

- The subtype training set is small and conditional on first-stage plant
  inclusion.
- Inputs longer than 512 tokens are truncated.
- The near-zero threshold reflects validation-set calibration on a small,
  imbalanced subtype dataset; it is not an intrinsic probability of technology
  focus.
- The model is trained on English title and abstract text, not full patent
  claims or descriptions.

## Reproducibility

Reference artifacts in the companion repository:

- Training report: `metadata/variety_technology_paecter_report_2026-01-30.md`
- Metrics JSON: `metadata/variety_technology_paecter_metrics_2026-01-30.json`
- Public companion repository:
  <https://github.com/aydiet/plant-patent-classifier-reproducibility>

## Companion Resources

- Original PaECTER base model:
  <https://huggingface.co/mpi-inno-comp/paecter>
- Interactive demo Space:
  <https://huggingface.co/spaces/aydiet/plant-patent-classifier-demo>
- First-stage plant-related detector:
  <https://huggingface.co/aydiet/plant-patent-paecter-plant-detector>
- Public companion repository:
  <https://github.com/aydiet/plant-patent-classifier-reproducibility>
- Manuscript context: *Identifying Plant-Related Patents: Corpus Construction
  and Global Patterns*

## Example

```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

model_id = "aydiet/plant-patent-paecter-variety-technology"
text = "Drought tolerant maize plant\n\nA maize plant with improved drought tolerance is provided."

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding="max_length", max_length=512)

with torch.no_grad():
    probs = torch.softmax(model(**inputs).logits, dim=-1)[0]

p_technology = float(probs[model.config.label2id["technology"]])
label = "technology" if p_technology >= 0.000399 else "variety"
print(label, p_technology)
```