FreqFault-tiny / README.md
bojian1's picture
Rename model repository and documentation to FreqFault
17fff10 verified
|
Raw History Blame Contribute Delete
1.14 kB
---
license: mit
tags:
- time-series
- fault-diagnosis
- pytorch
- safetensors
---
# FreqFault-tiny
Inference checkpoint for the **tiny** FreqFault time-frequency encoder.
| Item | Value |
| --- | ---: |
| Time embedding dimension | 64 |
| Frequency embedding dimension | 64 |
| Fused feature dimension | 128 |
| Attention heads | 2 |
| Layers per branch | 4 |
| Parameters | 420K |
| Weight size | 1.69 MB |
| Frequency bins | 800 |
| Input window | 0.1 seconds |
## Usage
```python
import numpy as np
import torch
from model import UniFreqProcessor, load_model
model, config = load_model("bojian1/FreqFault-tiny")
batch = UniFreqProcessor(config)([
np.random.randn(2048).astype("float32")
])
with torch.inference_mode():
features = model(batch)
```
Inference code and documentation:
https://github.com/Bojian-Chen/FreqFault
This repository contains inference artifacts only: `config.json` and
`model.safetensors`.
The repository was renamed from `bojian1/UniFreq-tiny`. Python API names
(`UniFreqProcessor`, `UniFreqModel`) and architecture identifiers remain
compatible with existing code. Model weights are unchanged.