Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
{}
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
# HyperSafe Deep Zero-Shot Classifier (ZSC)
|
| 6 |
+
|
| 7 |
+
## Model Description
|
| 8 |
+
This model is a custom Transformer-based Zero-Shot Classifier trained on the Ncert_dataset. It utilizes a DeepSafeEncoder architecture with memory-efficient gradient checkpointing.
|
| 9 |
+
|
| 10 |
+
### Technical Specifications
|
| 11 |
+
- **Architecture**: 12-layer Transformer Encoder
|
| 12 |
+
- **Embedding Dimension**: 256
|
| 13 |
+
- **Optimization**: AdamW with GradScaler (Mixed Precision)
|
| 14 |
+
- **Training Regime**: 2 Epochs via streaming (Zero-RAM impact)
|
| 15 |
+
|
| 16 |
+
## Performance Report (Evaluation Phase)
|
| 17 |
+
|
| 18 |
+
| Query | Predicted Label | Confidence Score | Result |
|
| 19 |
+
| :--- | :--- | :--- | :--- |
|
| 20 |
+
| What is a charge? | history | 0.9977 | ❌ False |
|
| 21 |
+
| The calculation of the derivative... | Advanced Mathematics | 0.9985 | ✅ True |
|
| 22 |
+
| How to bake a sourdough bread... | Software | 0.9987 | ❌ False |
|
| 23 |
+
| The solar system consists of... | Music | 0.9985 | ❌ False |
|
| 24 |
+
| The stock market saw a... | Art | 0.9979 | ❌ False |
|
| 25 |
+
| The Renaissance was a period... | History | 0.9982 | ✅ True |
|
| 26 |
+
| Basketball is played with... | Sports | 0.9972 | ✅ True |
|
| 27 |
+
| A balanced diet includes... | Nutrition | 0.9966 | ✅ True |
|
| 28 |
+
| The novel features a complex... | Literature | 0.9984 | ✅ True |
|
| 29 |
+
|
| 30 |
+
**Analysis**: The model shows strong potential for high-level semantic grouping but requires further training for fine-grained scientific categories.
|
| 31 |
+
|
| 32 |
+
## Usage Instructions
|
| 33 |
+
To use this model, download the weight file and load it into the `DeepSafeEncoder` architecture.
|
| 34 |
+
|
| 35 |
+
```python
|
| 36 |
+
import torch
|
| 37 |
+
# Define DeepSafeEncoder class
|
| 38 |
+
model = DeepSafeEncoder()
|
| 39 |
+
model.load_state_dict(torch.load('hyper_zsc_model.pt'))
|
| 40 |
+
model.eval()
|
| 41 |
+
```
|