File size: 4,374 Bytes
4500415
 
 
e243be0
 
936f740
e243be0
 
 
 
 
 
 
dc4db99
e243be0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
---
license: apache-2.0
---
<img src="https://huggingface.co/Vezora/Mistral-22B-v0.1/resolve/main/unsloth.png" width="100" height="150" />

### Mistral-22b-V.02 Release Announcement 🚀

## This model is not an moe, it is infact a 22B parameter dense model!

**Date**: April 11
**Creator** [Nicolas Mejia-Petit](https://twitter.com/mejia_petit)

### Overview
Just one day after the release of **Mixtral-8x-22b**, we are excited to introduce our handcrafted experimental model, **Mistral-22b-V.02**. This model is a culmination of equal knowledge distilled from all experts into a single, dense 22b model. This model is not a single trained expert, rather its a compressed MOE model, turning it into a dense 22b mode. This is the first working MOE to Dense model conversion.

### Capabilities
- **Math Proficiency**: The model exhibits strong mathematical abilities. Dispite not being trained on math.
- **Better at Coding** The model is significantly better at coding, than V1, it passed some of my simple coding test, such as "Create a simple HTML site with a button that changes the background color to a random color", which V1 failed.
- **More Cohesive** This V2 model is significantly more cohesive, and better at aunderstanding the prompts and answering with the appopriate answer.
- **Highly Uncencored** Since this model was also Re-Alligned to be uncencored, it can just answer anything you ask.
- **Multi Turn** The dataset this model trained on was mostly all multi turn conversations, spanning many different topics, with some emphasis on coding.
- **Json Mode** I did train this model on answering in JSON and using JSON tools., I have yet to try it, in depth but preliminary test shows it works, including.
- **Agent abilities** I did train this model on agent datasets, that teach it to do real world tasks such as picking up an object, and even navigating a webpage based off HTML.
- **Good Chili Recipe** The model gives a good chili recipe :)
- **16k Sequence Length** This model was trained with a 16k sequence length.

### Experimental Nature
Please note that Mistral-22b is still in a WIP. V.3 has started training now, with a different method than used before, this is to hopefully make the model more round in its internel knowlledge. Through my testing i found V2 to be a significant improvement over V.2, but I would like to do my testing with other methods.

### Upcoming Release: V.3
V-3 will feature a different base model for testing purposed, however this model is pretty darn good for a second test. :)

### Background
The decision to release this experimental version was prompted by someone attempting to replicate my experiment based on my tweets. We wanted to ensure our community has access to the official version first.

### Stay Updated
**V.3**, coming soon! And is currently training, will be done in the next ~24 hours. 🌟Paper Coming Soon🌟
- There will be more of these 22b models. They 5-6 siblings till I find what the best results are for MOE compression.
- However I am very surprised at how good this V.2 model is, off my small testing.

### Usage:
- This model requires a specific chat template, as the training format was Guanaco this is what it looks like:
- "### System: You are a helpful assistant. ### Human###: Give me the best chili recipe you can ###Assistant:"


## Thank you!
- Thank you to [Daniel Han](https://twitter.com/danielhanchen), for Unsloth AI which was used to train this model. this led to a 2-3x speed increae and 2-3x decrease in memmory consumption.
- Thank you to [Charles Coddard](https://twitter.com/chargoddard), for providng me with a script that was nessary to make this model.
- Thank you to Mistral, for releasing Another Wonderful open source model, under Apache 2.0.
- Thank you to [Tim Dettmers](https://twitter.com/Tim_Dettmers), for creating QLora
- Thank you to [Tri Dao](https://twitter.com/tri_dao), for creating Flash Attention
- Thank you to Microsoft, for the Lora paper, and the Slice-GPT paper.
- Thank you to the Hugging Face team, for everything.❤️ We really do appreciate you guys and all your hard work and commitment to the open source community!❤️


## Future plans, train 4-5 more of these expirmental models gather preliminary testing results, and then run evaluations on all the models I see have the best possibilities of excelling, then use the best one.