Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: transformers
|
| 4 |
+
---
|
| 5 |
+
# MidTrainingCheckpoint
|
| 6 |
+
|
| 7 |
+
<!-- markdownlint-disable first-line-h1 -->
|
| 8 |
+
<!-- markdownlint-disable html -->
|
| 9 |
+
<!-- markdownlint-disable no-duplicate-header -->
|
| 10 |
+
|
| 11 |
+
<div align="center">
|
| 12 |
+
<img src="figures/fig1.png" width="60%" alt="MidTrainingCheckpoint" />
|
| 13 |
+
</div>
|
| 14 |
+
<hr>
|
| 15 |
+
|
| 16 |
+
## 1. Introduction
|
| 17 |
+
|
| 18 |
+
MidTrainingCheckpoint is a snapshot taken at the midpoint of our training run. It captures the model state at step 500, providing a useful reference for studying training dynamics.
|
| 19 |
+
|
| 20 |
+
<p align="center">
|
| 21 |
+
<img width="80%" src="figures/fig3.png">
|
| 22 |
+
</p>
|
| 23 |
+
|
| 24 |
+
This checkpoint is particularly useful for:
|
| 25 |
+
- Comparing with earlier and later checkpoints
|
| 26 |
+
- Understanding the training trajectory
|
| 27 |
+
- Performing intermediate model analysis
|
| 28 |
+
|
| 29 |
+
## 2. Model Details
|
| 30 |
+
|
| 31 |
+
| Property | Value |
|
| 32 |
+
|---|---|
|
| 33 |
+
| Architecture | BERT |
|
| 34 |
+
| Training Steps | 500 |
|
| 35 |
+
| Checkpoint Name | step_500 |
|
| 36 |
+
| Purpose | Mid-training reference |
|
| 37 |
+
|
| 38 |
+
## 3. Usage
|
| 39 |
+
|
| 40 |
+
```python
|
| 41 |
+
from transformers import AutoModel
|
| 42 |
+
|
| 43 |
+
model = AutoModel.from_pretrained("MidTraining-Checkpoint")
|
| 44 |
+
```
|
| 45 |
+
|
| 46 |
+
## 4. License
|
| 47 |
+
[MIT License](LICENSE)
|
| 48 |
+
|
| 49 |
+
## 5. Contact
|
| 50 |
+
Open an issue on GitHub.
|