dleemiller commited on
Commit
d7e4055
·
verified ·
1 Parent(s): bc08380

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -3
README.md CHANGED
@@ -19,18 +19,24 @@ metrics:
19
 
20
  # SwipeALot Base Model
21
 
22
- Multimodal, multi-objective transformer for swipe keyboard prediction. Trained on the [futo-org/swipe.futo.org](https://huggingface.co/datasets/futo-org/swipe.futo.org) dataset.
 
23
 
24
  This model is trained with the following objectives:
25
- - Masked character prediction
26
  - Masked path prediction
27
  - Text length prediction (CLS token)
28
- - Path/text embedding (SEP token)
29
 
30
  <p align="center">
31
  <img src="https://cdn-uploads.huggingface.co/production/uploads/65ff92ea467d83751a727538/OV87xy-_ID0TqKW0bfvVq.png" style="width: 400px;">
32
  </p>
33
 
 
 
 
 
 
34
  ## Quick Start
35
 
36
  ```python
 
19
 
20
  # SwipeALot Base Model
21
 
22
+ Multimodal, multi-objective transformer for swipe keyboard prediction.
23
+ Trained on the [futo-org/swipe.futo.org](https://huggingface.co/datasets/futo-org/swipe.futo.org) dataset.
24
 
25
  This model is trained with the following objectives:
26
+ - Masked character prediction (MLM)
27
  - Masked path prediction
28
  - Text length prediction (CLS token)
29
+ - Path/text embedding (SEP token, contrastive + Matryoshka@ 64, 128, 384, 768)
30
 
31
  <p align="center">
32
  <img src="https://cdn-uploads.huggingface.co/production/uploads/65ff92ea467d83751a727538/OV87xy-_ID0TqKW0bfvVq.png" style="width: 400px;">
33
  </p>
34
 
35
+
36
+ > [!NOTE]
37
+ > This model should be further fine-tuned for a specific task, if not using the embedding mode.
38
+ > For example, length prediction can be significantly improved in a single task setting.
39
+
40
  ## Quick Start
41
 
42
  ```python