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 |
|