File size: 3,574 Bytes
5fae50c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9f3ac5c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5fae50c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
---
library_name: pytorch
tags:
- contrastive-learning
- tag-classification
- semantic-search
- embeddings
- persona-conditioned
- pretrained-backbone
---

# modernbert-base-tag-classification

This is a **pretrained backbone model** (answerdotai/ModernBERT-base) used for tag classification via contrastive learning.

## Model Description

This model uses the `answerdotai/ModernBERT-base` backbone directly without fine-tuning. It's designed for zero-shot tag classification tasks where you want to use a pretrained embedding model for semantic similarity computation.

## Usage

See the README.md for detailed usage examples using our module abstractions.

## Model Architecture

- **Backbone**: `answerdotai/ModernBERT-base`
- **Type**: Pretrained backbone (no fine-tuning)
- **Embedding Dimension**: Varies by backbone model

## Usage Example

```python
"""
Example: Using ModernBERT-base for Tag Classification

This example shows how to use the pretrained ModernBERT-base backbone
for zero-shot tag classification using our module abstractions.

Installation:
    pip install git+https://github.com/Pieces/TAG-module.git@main
    # Or: pip install -e .

Note: ModernBERT requires Python < 3.14 due to torch.compile compatibility.
"""

import torch
from tags_model.models.backbone import SharedTextBackbone
from playground.validate_from_checkpoint import compute_ranked_tags

# Load the pretrained backbone
print("Loading ModernBERT-base...")
backbone = SharedTextBackbone(
    model_name="answerdotai/ModernBERT-base",
    embedding_dim=768,
    freeze_backbone=True,
    pooling_mode="cls",
    trust_remote_code=True,  # Required for ModernBERT
)
backbone.eval()
print("✓ Model loaded!")

# Example query
query_text = "Machine learning model for image classification using PyTorch"

# Candidate tags to rank
candidate_tags = [
    "pytorch", "machine-learning", "deep-learning", "computer-vision",
    "neural-networks", "cnn", "image-classification", "tensorflow",
    "data-science", "python"
]

print(f"\nQuery: {query_text}")
print(f"Candidate tags: {candidate_tags}\n")

# Encode query and tags
with torch.inference_mode():
    query_emb = backbone.encode_texts([query_text], max_length=512, return_dict=False)[0]
    tag_embs = backbone.encode_texts(candidate_tags, max_length=512, return_dict=False)

print(f"Query embedding shape: {query_emb.shape}")
print(f"Tag embeddings shape: {tag_embs.shape}")

# Rank tags by similarity
ranked_tags = compute_ranked_tags(
    query_emb=query_emb,
    pos_embs=torch.empty(0, 768),  # No positives for zero-shot
    neg_embs=torch.empty(0, 768),  # No negatives for zero-shot
    general_embs=tag_embs,
    positive_tags=[],
    negative_tags=[],
    general_tags=candidate_tags,
)

# Display top-ranked tags
print("\n" + "="*60)
print("Top Ranked Tags:")
print("="*60)
for tag, rank, label, score in ranked_tags[:5]:
    print(f"{rank:2d}. {tag:20s} (score: {score:.4f})")

print("\n" + "="*60)
print("Example complete!")


```

### Running the Example

```bash
# Install the repository first
pip install git+https://github.com/Pieces/TAG-module.git@main
# Or for local development:
pip install -e .

# Run the example
python modernbert_example.py
```


## Citation

If you use this model, please cite:

```bibtex
@software{{tag_module,
  title = {{TAG Module: Persona-Conditioned Contrastive Learning for Tag Classification}},
  author = {{Your Name}},
  year = {{2025}},
  url = {{https://github.com/yourusername/tag-module}}
}}
```

## License

Please refer to the original model license for the backbone model.