docs: note sub_10 only training following chayan
Browse files
README.md
CHANGED
|
@@ -185,7 +185,9 @@ checkpoints/
|
|
| 185 |
|
| 186 |
## Training Details
|
| 187 |
|
| 188 |
-
-
|
|
|
|
|
|
|
| 189 |
- **Method**: Nearest-neighbor regression with cosine distance, distance-weighted
|
| 190 |
- **Evaluation**: Full 8,400 RouterArena queries (no data leakage)
|
| 191 |
- **Training Time**: < 1 second
|
|
|
|
| 185 |
|
| 186 |
## Training Details
|
| 187 |
|
| 188 |
+
Following [chayan](https://huggingface.co/adaptive-classifier/chayan), we only use the official **sub_10 split** (809 queries, 10% of the full 8,400) for training. No full-set data is used during training or hyperparameter tuning.
|
| 189 |
+
|
| 190 |
+
- **Training Data**: RouterArena sub_10 split (809 queries)
|
| 191 |
- **Method**: Nearest-neighbor regression with cosine distance, distance-weighted
|
| 192 |
- **Evaluation**: Full 8,400 RouterArena queries (no data leakage)
|
| 193 |
- **Training Time**: < 1 second
|