Update README.md
Browse files
README.md
CHANGED
|
@@ -70,6 +70,9 @@ The format for TinyCoT was:
|
|
| 70 |
|
| 71 |
Memphis outperforms human-data models that are over twice its size, along with SFT models of its size, and trades with the Zephyr DPO model. That said, Zephyr uses synthetic data, and *much* more of it.
|
| 72 |
|
|
|
|
|
|
|
|
|
|
| 73 |
It is unclear why Zephyr performs so poorly on BBH. Perhaps it is overfit.
|
| 74 |
|
| 75 |
Notes:
|
|
|
|
| 70 |
|
| 71 |
Memphis outperforms human-data models that are over twice its size, along with SFT models of its size, and trades with the Zephyr DPO model. That said, Zephyr uses synthetic data, and *much* more of it.
|
| 72 |
|
| 73 |
+
Note that BBH results have wide SEs, exceeding 16%.
|
| 74 |
+
|
| 75 |
+
|
| 76 |
It is unclear why Zephyr performs so poorly on BBH. Perhaps it is overfit.
|
| 77 |
|
| 78 |
Notes:
|