File size: 1,645 Bytes
31104bf
 
 
 
 
 
 
 
 
96fd405
77f5715
8ccae35
f2afbbe
0f83c3e
e22322c
fa26741
5378b3c
 
 
 
 
 
 
 
 
fa26741
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
---
title: README
emoji: πŸš€
colorFrom: red
colorTo: purple
sdk: static
pinned: false
---

<img src="https://media.discordapp.net/attachments/1074346695191711875/1090859826793431139/image.png?width=1031&height=152" width=100%\>

Who needs em, we all have em, they're just like us. Unusable models, compute optimally πŸ”₯. We hope that by open-sourcing our compute-optimal trained models, that others can replicate our results and also make no use out of our unusable models. These models are not useful in the slightest, and don't benefit research. Every time you use one of these models, you can be sure that you will not get a useful result, and every time we kiss I swear I can fly. Can't you feel my heart beat fast, I want this to last, need you by my side. We introduce a cascade(a) (sorry) of classes and models.

Evaluations and more information about the training for every Gerbil model and the mixture-of-tasks Blender pretraining method inspired by UL2 can be found here: https://github.com/aicrumb/notebook-hosting/blob/main/GerbilLabEvaluations.md

Special tokens for "Blender" models' pretraining include:

```
'<fitm_start>', '<multiple_tok_mask>', '<fitm_result>', '<causal>', '<mlm_start>', '<single_tok_mask>', '<mlm_end>'
# Example fill in the middle
'<fitm_start> this is an <multiple_tok_mask> for fill-in-the-middle <fitm_result> example text <|endoftext|>'
# Example causal language modelling
'<causal> this is an example text for causal language modelling <|endoftext|>'
# Example masked language modelling
'<mlm_start> this is an <single_tok_mask> text for masked language modelling <mlm_end> example <|endoftext|>'
```