Downloads · 30 days
6
8% of all-time downloads
osmosis-modernbert-joint
KingTechnician/osmosis-modernbert-joint
osmosis-modernbert-joint is a text classification model from KingTechnician. Use it when you need a label for a piece of text. It is set up for sentence-transformers. The card lists the license as mit.
Five-class response sufficiency classifier trained with SetFit-style contrastive learning on Yahoo Answers + Triage data.
All-time downloads
80
Public
Parameters
149M
596 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors596 MB · 99%
At a glance
- Task
- Text Classification
- Library
- sentence-transformers
- License
- mit
- Model type
- modernbert
- Access
- Public
- Created
- Apr 14, 2026
- Updated
- Apr 14, 2026
- SHA
- a2878f8d
Try a prompt
Base models
Model card
From the Hugging Face model README
OSMoSIS Joint Classifier (ModernBERT-embed-base)
Five-class response sufficiency classifier trained with SetFit-style contrastive learning on Yahoo Answers + Triage data.
Performance
| Evaluation | Accuracy | Macro F1 |
|---|---|---|
| Yahoo within-domain | 51.9% | 41.3% |
| Triage held-out | 94.7% | 94.7% |
Per-Class F1 (Triage held-out)
| Class | F1 |
|---|---|
| ADDR_DIRECT | 93.0% |
| ADDR_PARTIAL | 94.2% |
| NOADDR_ON | 93.6% |
| NOADDR_TANGENTIAL | 92.5% |
| NOADDR_OFF | 100% |
Usage
from sentence_transformers import SentenceTransformer
import joblib
encoder = SentenceTransformer("KingTechnician/osmosis-modernbert-joint")
head = joblib.load("head.joblib") # download from this repo
text = "Objective:\nWhat causes rain?\nResponse: Rain forms when water vapor condenses."
embedding = encoder.encode([text])
prediction = head.predict(embedding)
labels = ["ADDR_DIRECT", "ADDR_PARTIAL", "NOADDR_ON", "NOADDR_TANGENTIAL", "NOADDR_OFF"]
print(labels[prediction[0]])
Training
- Method: Contrastive fine-tuning (20K class-balanced pairs, 3 epochs) + Logistic Regression head
- Data: Yahoo Answers OSMoSIS (24,890) + Triage synthetic (1,492)
- Base model: nomic-ai/modernbert-embed-base (149M params)
- Task
- Text Classification
- Library
- sentence-transformers
- Type
- modernbert
- License
- mit
- Languages
- en
- Created
- Apr 14, 2026
- Updated
- Apr 14, 2026