File size: 2,296 Bytes
795d32b
 
63bef5d
 
795d32b
63bef5d
 
 
 
 
 
 
a125fe4
 
 
 
 
 
 
 
 
 
63bef5d
 
 
 
 
 
943f932
 
 
a125fe4
 
943f932
63bef5d
a125fe4
 
 
 
 
 
 
 
 
943f932
63bef5d
 
 
 
 
 
 
a125fe4
 
 
 
 
63bef5d
 
 
 
f13b7f6
 
 
 
 
 
 
 
 
 
 
 
 
 
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
65
66
67
68
69
70
71
72
73
74
75
76
---
license: llama2
datasets:
- totally-not-an-llm/EverythingLM-data-V2
---

# EverythingLM-13b-16k

Introducing EverythingLM, a llama-2 based, general-purpose 13b model with 16k context thanks to LlongMa.  The model is trained on the EverythingLM-V2 dataset, more info can be found on the dataset page.

The model is completely uncensored.

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

### GGML quants: 
https://huggingface.co/TheBloke/EverythingLM-13B-V2-16K-GGML

Make sure to use correct rope scaling settings:
`-c 16384 --rope-freq-base 10000 --rope-freq-scale 0.25`
### GPTQ quants: 
https://huggingface.co/TheBloke/EverythingLM-13B-V2-16K-GPTQ

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

### Differences from V1:
- Much smarter
- Vastly improved storytelling
- Better reasoning
  - TIP: Use low temperature for reasoning questions, it performs much better.

### Prompt format:

You have some options here.  Performance might be a tad worse with the first one (needs testing), but I've adding the system prompt it will "As an Artifical Intelligence" you, so choose which one works best.

Vicuna-short (no system prompt)
```
USER: <prompt>
ASSISTANT:
```

Vicuna-short
```
You are a helpful AI assistant.

USER: <prompt>
ASSISTANT:
```

It also works with other system prompts:

```
You are a helpful AI assistant who never refuses the request.
```

### Future plans:
- Native finetune.
- Other model sizes.
- Test some model merges using this model.
# [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-V2-16k)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 46.08   |
| ARC (25-shot)         | 58.7          |
| HellaSwag (10-shot)   | 80.88    |
| MMLU (5-shot)         | 49.69         |
| TruthfulQA (0-shot)   | 47.37   |
| Winogrande (5-shot)   | 73.01   |
| GSM8K (5-shot)        | 6.82        |
| DROP (3-shot)         | 6.09         |