Hugging Face's logo

inflatebot
/
L3-8B-Helium3

Text Generation
Transformers
Safetensors
llama
mergekit
Merge
Not-For-All-Audiences
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
L3-8B-Helium3
File size: 2,452 Bytes 
de3728e
 
 
 
 
 
 
 
 
 
 
 
747a99d
de3728e
 
 
 
747a99d
 
 
 
 
b8e6057
747a99d
b8e6057
747a99d
b8e6057
 
747a99d
 
 
 
 
 
 
 
 
de3728e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
57
58
59
60
61
62
63
64
 
---
base_model:
- inflatebot/helide-beta-r3
- inflatebot/helide-beta-r1
- inflatebot/helide-beta-r4
- inflatebot/helide-beta-r0
library_name: transformers
tags:
- mergekit
- merge

---
# L3-Helium3-8B

This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).

## Merge Details

There was a problem with the Helide beta. 3 models resulted, each of which had different strengths. But they came about as a result of balancing two models.
That math wasn't quite mathing. There wasn't going to be a way to get the best of all three worlds just by tweaking a SLERP ratio.

But there were three of them.

The name was serendipity.

The layup was obscene.

But I *live* for the bit.

Helium-3 is a RP and storywriting hybrid, ultimately based on Sao10K's [Stheno](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2) and Fizzarolli's [Rosier](https://huggingface.co/Fizzarolli/L3-8b-Rosier-v1), and the culmination of the Helide project.
Combining Rosier's prose and knowledge of niche fetish with Stheno's steerability and crackling personality, Helium-3 brings the advancements of modern AI models to the Freaks™.
They'll chew you up and spit you out just as readily as they'll shower you with affection.

I'm genuinely proud of this one. This is the model I wish existed.

Thank you to [Fizzarolli](https://huggingface.co/Fizzarolli) for consulting and providing technical assistance which accelerated the second leg of this project from several weeks into a single night, and for making the Rosier model that made this possible. On several levels, H3 wouldn't have been possible without her.

### Merge Method

This model was merged using the [Model Stock](https://arxiv.org/abs/2403.19522) merge method using [inflatebot/helide-beta-r1](https://huggingface.co/inflatebot/helide-beta-r1) as a base.

### Models Merged

The following models were included in the merge:
* [inflatebot/helide-beta-r3](https://huggingface.co/inflatebot/helide-beta-r3)
* [inflatebot/helide-beta-r4](https://huggingface.co/inflatebot/helide-beta-r4)
* [inflatebot/helide-beta-r0](https://huggingface.co/inflatebot/helide-beta-r0)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
models:
  - model: inflatebot/helide-beta-r4
  - model: inflatebot/helide-beta-r0
  - model: inflatebot/helide-beta-r3

merge_method: model_stock
base_model: inflatebot/helide-beta-r1
dtype: bfloat16

```