readme: sftorm hyperparams

readme.qmd

@@ -74,11 +74,18 @@ We guess the SFT orm just needs more data.
 This aligns with [@lightman2023let]'s note that 2 epochs improve performance on smaller datasets.
 More epochs don't help much after a point, especially on larger datasets.
 
+In my experiments, I follow the Llamafactory example by setting the epoch to 3.
+I discover that a smaller learning rate of 1e-6 yields better results.
+The batch size is configured as 2 (per device) multiplied by 4 (device number).
+Under this setup, the performance of SFTorm is on par with ORM, albeit slightly lower than the ORM results.
+
 | model          | interview | competition | introductory |
-| :---:          | :-------: | :---------: | :-----------:|
+| :---:          | :-------: | :---------: | :----------: |
 | random         | 21.4%     | 8.7%        | 34.4%        |
+| orm(max)       | 53.8%     | 27.2%       | 50%          |
+| orm(min)       | 59.6%     | 27.2%       | 57.7%        |
 | sftorm(epoch=3)| 36.5%     | 27.2%       | 42.3%        |
-| orm            | 53.8%     | 27.2%       | 50%          |
+| sftorm(lr=1e-6)| 51.9%     | 27.2%       | 53.8%        |
 
 ## Environment