AmelieSchreiber
commited on
Commit
•
45e45d6
1
Parent(s):
c20ef98
Update README.md
Browse files
README.md
CHANGED
@@ -10,7 +10,16 @@ tags:
|
|
10 |
- biology
|
11 |
---
|
12 |
|
|
|
|
|
13 |
These are the checkpoints for the first ever QLoRA for ESM-2! They haven't been checked for overfitting yet, so use with caution!
|
14 |
You can load and use them similarly to the LoRA models. This is the smallest `esm2_t6_8M_UR50D` model, so the metrics aren't great.
|
15 |
Scaling to larger models for better metrics is in progress.
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
- biology
|
11 |
---
|
12 |
|
13 |
+
# ESM-2 QLoRA
|
14 |
+
|
15 |
These are the checkpoints for the first ever QLoRA for ESM-2! They haven't been checked for overfitting yet, so use with caution!
|
16 |
You can load and use them similarly to the LoRA models. This is the smallest `esm2_t6_8M_UR50D` model, so the metrics aren't great.
|
17 |
Scaling to larger models for better metrics is in progress.
|
18 |
|
19 |
+
## QLoRA Info
|
20 |
+
|
21 |
+
Note, we are only training 0.58% of the parameters, using only the query, key, and value weight matrices.
|
22 |
+
|
23 |
+
```
|
24 |
+
trainable params: 23682 || all params: 4075265 || trainable%: 0.5811155838945443
|
25 |
+
```
|