Create README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,88 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- microsoft/deberta-v3-large
|
| 5 |
+
---
|
| 6 |
+
Multi-Task Product and Hazard Classifier
|
| 7 |
+
This model performs multi-task classification to predict both product categories and hazard categories from text descriptions. It's based on DeBERTa-v3 architecture and trained to identify product types and potential hazards simultaneously.
|
| 8 |
+
Model Description
|
| 9 |
+
|
| 10 |
+
Model Type: Multi-task classification (DeBERTa-v3 base)
|
| 11 |
+
Languages: English
|
| 12 |
+
Pipeline Tag: text-classification
|
| 13 |
+
Max Sequence Length: 1024 tokens
|
| 14 |
+
|
| 15 |
+
Usage
|
| 16 |
+
pythonCopyfrom transformers import AutoTokenizer, AutoModel
|
| 17 |
+
import torch
|
| 18 |
+
from torch.nn import functional as F
|
| 19 |
+
|
| 20 |
+
# Load model and tokenizer
|
| 21 |
+
tokenizer = AutoTokenizer.from_pretrained("your-username/model-name")
|
| 22 |
+
model = AutoModel.from_pretrained("your-username/model-name")
|
| 23 |
+
|
| 24 |
+
# Prepare your text
|
| 25 |
+
text = "Your product description here"
|
| 26 |
+
|
| 27 |
+
# Tokenize and prepare input
|
| 28 |
+
inputs = tokenizer(
|
| 29 |
+
text,
|
| 30 |
+
padding=True,
|
| 31 |
+
truncation=True,
|
| 32 |
+
max_length=1024,
|
| 33 |
+
return_tensors="pt",
|
| 34 |
+
return_token_type_ids=False
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
+
# Run inference
|
| 38 |
+
with torch.no_grad():
|
| 39 |
+
outputs = model(**inputs)
|
| 40 |
+
product_logits = outputs['product_logits']
|
| 41 |
+
hazard_logits = outputs['hazard_logits']
|
| 42 |
+
|
| 43 |
+
product_probs = F.softmax(product_logits, dim=-1)
|
| 44 |
+
hazard_probs = F.softmax(hazard_logits, dim=-1)
|
| 45 |
+
|
| 46 |
+
# Get predictions
|
| 47 |
+
product_predictions = product_probs.cpu().numpy()
|
| 48 |
+
hazard_predictions = hazard_probs.cpu().numpy()
|
| 49 |
+
Prediction Labels
|
| 50 |
+
Product Categories
|
| 51 |
+
pythonCopyproduct_labels = {
|
| 52 |
+
'0': 'label_0',
|
| 53 |
+
'1': 'label_1',
|
| 54 |
+
# Add your product category labels here
|
| 55 |
+
}
|
| 56 |
+
Hazard Categories
|
| 57 |
+
pythonCopyhazard_labels = {
|
| 58 |
+
'0': 'label_0',
|
| 59 |
+
'1': 'label_1',
|
| 60 |
+
# Add your hazard category labels here
|
| 61 |
+
}
|
| 62 |
+
Model Limitations
|
| 63 |
+
|
| 64 |
+
The model is designed for English text only
|
| 65 |
+
Maximum input length is 1024 tokens
|
| 66 |
+
Performance may vary for texts significantly different from the training data
|
| 67 |
+
|
| 68 |
+
Training Data
|
| 69 |
+
The model was trained on a dataset containing product descriptions with their corresponding product categories and hazard classifications. The training data includes various product types and potential hazards commonly found in consumer products.
|
| 70 |
+
Evaluation Results
|
| 71 |
+
[Add your model's evaluation metrics here]
|
| 72 |
+
Intended Uses & Limitations
|
| 73 |
+
Intended Uses:
|
| 74 |
+
|
| 75 |
+
Product categorization
|
| 76 |
+
Hazard identification in product descriptions
|
| 77 |
+
Safety analysis of product text
|
| 78 |
+
|
| 79 |
+
Limitations:
|
| 80 |
+
|
| 81 |
+
Should not be used as the sole decision maker for safety-critical applications
|
| 82 |
+
Requires human verification for important safety decisions
|
| 83 |
+
May not recognize new or unusual product types/hazards
|
| 84 |
+
|
| 85 |
+
Citation
|
| 86 |
+
[Add citation information if applicable]
|
| 87 |
+
Contact
|
| 88 |
+
[Your contact information or where to report issues]
|