Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
# 萬題v1 (WanTopic-V1-Minimal)
|
| 3 |
+
|
| 4 |
+
## Model Description
|
| 5 |
+
- **Author**: ShihteSiao
|
| 6 |
+
- **Version**: v1.4-minimal
|
| 7 |
+
- **Architecture**: Hierarchical Reasoning Model (HRM) - Minimal 256d
|
| 8 |
+
- **Parameters**: 5,552,386 (5.6M)
|
| 9 |
+
- **Memory Usage**: 0.0GB
|
| 10 |
+
|
| 11 |
+
## Performance
|
| 12 |
+
- **Test Accuracy**: 50.0%
|
| 13 |
+
- **Best Validation Loss**: nan
|
| 14 |
+
- **Peak Reasoning Efficiency**: 79.2%
|
| 15 |
+
- **Max Reasoning Steps**: 12
|
| 16 |
+
|
| 17 |
+
## Architecture Details
|
| 18 |
+
- **Hidden Size**: 256
|
| 19 |
+
- **Attention Heads**: 4
|
| 20 |
+
- **High-level Layers**: 4
|
| 21 |
+
- **Low-level Layers**: 4
|
| 22 |
+
- **Total Layers**: 8
|
| 23 |
+
|
| 24 |
+
## Training Configuration
|
| 25 |
+
- **Batch Size**: 4 × 8 = 32
|
| 26 |
+
- **Learning Rate**: 0.0002
|
| 27 |
+
- **Epochs**: 40
|
| 28 |
+
- **Mixed Precision**: True
|
| 29 |
+
- **Label Smoothing**: 0.0
|
| 30 |
+
|
| 31 |
+
## Usage
|
| 32 |
+
```python
|
| 33 |
+
import torch
|
| 34 |
+
|
| 35 |
+
# Load model
|
| 36 |
+
checkpoint = torch.load('WanTopic-V1-Stable_final_model.pth')
|
| 37 |
+
# Initialize model with checkpoint['model_config']
|
| 38 |
+
# Load state dict: model.load_state_dict(checkpoint['model_state_dict'])
|
| 39 |
+
```
|
| 40 |
+
|
| 41 |
+
Generated on: 2025-09-01 14:03:02
|