File size: 2,078 Bytes
601e5a5
cb4ca0d
 
 
601e5a5
cb4ca0d
 
 
 
 
 
 
 
 
a78c44a
 
1de9244
a78c44a
1de9244
a78c44a
 
cb4ca0d
 
 
 
 
 
 
f880c5c
97a842d
d338140
cb4ca0d
 
 
 
 
 
 
 
 
 
 
 
2a47555
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: llama2
datasets:
- totally-not-an-llm/EverythingLM-data-V3
---

# EverythingLM-13b-V3-16k

Introducing EverythingLM, a llama-2 based, general-purpose 13b model with 16k context thanks to LlongMa.  The model is trained on the EverythingLM-V3 dataset, more info can be found on the dataset page.

The model is completely uncensored.

Despite being "uncensored", the base model might be resistant; you might have to prompt-engineer certain prompts.

### Quants (Thanks TheBloke!):
https://huggingface.co/TheBloke/EverythingLM-13B-V3-16K-GGUF

https://huggingface.co/TheBloke/EverythingLM-13B-V3-16K-GPTQ

https://huggingface.co/TheBloke/EverythingLM-13B-V3-16K-AWQ

### Notable features:
- Automatically triggered CoT reasoning.
- Verbose and detailed replies.
- Creative stories.
- Good prompt understanding.

### Differences from V2:
- Much more uncensored.
- Actual roleplaying ability now!
- General all around improvements thanks to the new dataset.  Check out the dataset for more info.

### Prompt format (Alpaca-chat):

```
USER: <prompt>
ASSISTANT:
```

### Future plans:
- Highest priority right now is V3.1 with more optimized training and iterative dataset improvements based on testing.

### Note:
Through testing V2, I realized some alignment data had leaked in, causing the model to be less cooperative then intended.  This model should do much better due to stricter filetering.
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_totally-not-an-llm__EverythingLM-13b-V3-16k)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 44.82   |
| ARC (25-shot)         | 58.19          |
| HellaSwag (10-shot)   | 80.12    |
| MMLU (5-shot)         | 50.48         |
| TruthfulQA (0-shot)   | 45.18   |
| Winogrande (5-shot)   | 70.72   |
| GSM8K (5-shot)        | 1.97        |
| DROP (3-shot)         | 7.06         |