Update README.md
Browse files
README.md
CHANGED
@@ -2,6 +2,9 @@
|
|
2 |
license: mit
|
3 |
---
|
4 |
|
|
|
|
|
|
|
5 |
#datasets
|
6 |
|
7 |
- Yale-LILY/FOLIO
|
@@ -146,3 +149,15 @@ training settings
|
|
146 |
max_grad_norm=1.0,
|
147 |
lr_scheduler_type='cosine',
|
148 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: mit
|
3 |
---
|
4 |
|
5 |
+
code to train
|
6 |
+
- https://gist.github.com/thistleknot/398466a4bd75a1315825d7f04db635a6
|
7 |
+
|
8 |
#datasets
|
9 |
|
10 |
- Yale-LILY/FOLIO
|
|
|
149 |
max_grad_norm=1.0,
|
150 |
lr_scheduler_type='cosine',
|
151 |
```
|
152 |
+
|
153 |
+
Evaluation:
|
154 |
+
|
155 |
+
I kept tweaking the model until I got confirmations from chatgpt 4, but the final training error (1 epoch) came in under .5 at the end.
|
156 |
+
|
157 |
+
:tell me something I don't know</s> :exists x1.(_thing(x1) & _donknow(x1))
|
158 |
+
Does the fol expression fit?
|
159 |
+
|
160 |
+
Depending on how I asked (for example, it would almost always suggest revisions if I asked
|
161 |
+
- Is the fol expression adequate?
|
162 |
+
- How faithful is the fol expression?
|
163 |
+
)
|