Update README.md
Browse files
README.md
CHANGED
|
@@ -160,7 +160,7 @@ scores = (embeddings[:2] @ embeddings[2:].T)
|
|
| 160 |
```
|
| 161 |
### FAQs
|
| 162 |
**1. Does the model support MRL?**<br>
|
| 163 |
-
The model currently does not support MRL in this release due to observed performance degradation
|
| 164 |
**2. Why not build upon the Qwen3 series models?**<br>
|
| 165 |
Our initial research experiments commenced prior to the release of Qwen3. To maintain our experimental consistency, we retained the original base model throughout the study. While we subsequently conducted first-stage (retrieval) training with Qwen3, the performance after 32k steps showed no significant improvement over Qwen2.5, leading to discontinuation of further development with this architecture.
|
| 166 |
|
|
|
|
| 160 |
```
|
| 161 |
### FAQs
|
| 162 |
**1. Does the model support MRL?**<br>
|
| 163 |
+
The model currently does not support MRL in this release due to observed performance degradation.<br>
|
| 164 |
**2. Why not build upon the Qwen3 series models?**<br>
|
| 165 |
Our initial research experiments commenced prior to the release of Qwen3. To maintain our experimental consistency, we retained the original base model throughout the study. While we subsequently conducted first-stage (retrieval) training with Qwen3, the performance after 32k steps showed no significant improvement over Qwen2.5, leading to discontinuation of further development with this architecture.
|
| 166 |
|