Remove aggregate R2 leaderboard from model card
Browse files
README.md
CHANGED
|
@@ -27,22 +27,6 @@ The model is trained **entirely on synthetic data**.
|
|
| 27 |
|
| 28 |
## Results
|
| 29 |
|
| 30 |
-
Regression performance across **96 evaluation tasks** from 3 benchmark sources
|
| 31 |
-
(TabArena, TALENT, OpenML-Reg):
|
| 32 |
-
|
| 33 |
-
| Rank | Model | Mean R² | Training Data |
|
| 34 |
-
|---|---|---|---|
|
| 35 |
-
| **1** | **Synthefy Tabular + Thinking** | **0.7487** | Synthetic |
|
| 36 |
-
| **2** | **Synthefy Tabular** | **0.7475** | Synthetic |
|
| 37 |
-
| 3 | TabPFN-3 | 0.7443 | Synthetic |
|
| 38 |
-
| 4 | TabPFN-2.6 | 0.7437 | Synthetic |
|
| 39 |
-
| 5 | Real TabPFN-2.5 | 0.7364 | Real + Synthetic |
|
| 40 |
-
| 6 | TabICLv2 | 0.7354 | Synthetic |
|
| 41 |
-
| 7 | TabPFN-2.5 | 0.7354 | Synthetic |
|
| 42 |
-
| 8 | LimiX-2M (pretrained baseline) | 0.7301 | Synthetic |
|
| 43 |
-
|
| 44 |
-
Synthefy Tabular is **first on aggregate mean R²**, ahead of TabPFN-3, at ~5.5M parameters.
|
| 45 |
-
|
| 46 |
### By source
|
| 47 |
|
| 48 |
| Source | N | Synthefy Tabular (best) | TabPFN-3 | Result |
|
|
|
|
| 27 |
|
| 28 |
## Results
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
### By source
|
| 31 |
|
| 32 |
| Source | N | Synthefy Tabular (best) | TabPFN-3 | Result |
|