97 lines
3.2 KiB
Markdown
97 lines
3.2 KiB
Markdown
---
|
|
license: cc-by-nc-4.0
|
|
---
|
|
|
|
MLewd is a model created to be... Lewd. That's all. Based on ReMM.
|
|
|
|
There was so much attempt on this model that I can't count them all. Bear with me lmao.
|
|
|
|
The OG plan: https://pastebin.com/hfJ80rKL
|
|
|
|
Command useds and explaination :
|
|
```shell
|
|
Due to hardware limitation, some merge was done in 2 part.
|
|
|
|
Last mix :
|
|
- ReMM (Base) (0.57)
|
|
- Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged (Llama Chat Uncensored) (0.35)
|
|
- KoboldAI/LLAMA2-13B-Holodeck-1 (0.08)
|
|
|
|
Part 1: python ties_merge.py TheBloke/Llama-2-13B-fp16 ./MLewdBase-L2-13B-part1 --merge Undi95/ReMM-L2-13B --density 0.88 --merge KoboldAI/LLAMA2-13B-Holodeck-1 --density 0.12 --cuda
|
|
|
|
Part 2: python ties_merge.py TheBloke/Llama-2-13B-fp16 ./MLewdBase-L2-13B --merge Undi95/MLewdBase-L2-13B-part1 --density 0.65 --merge Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged --density 0.35 --cuda
|
|
|
|
(MLewd-L2-13B-v1-2 got disqualified)
|
|
|
|
- Applying LoRA: nRuaif/Kimiko-v2-13B at (0.24) weight on MLewd-L2-13B-v1-1
|
|
=> Result: MLewd-L2-13B-v1-3
|
|
|
|
================== ERP RANKING TEST ===========================
|
|
|
|
19.42 | MLewd-L2-13B-v1-3.q5_K_M.gguf (-> Best)
|
|
19.25 | MLewd-L2-13B-v1-1.q5_K_M.gguf
|
|
18.25 | MLewd-L2-13B-v1-2.q5 K M.gguf
|
|
|
|
================== RETRY ===========================
|
|
|
|
Mix:
|
|
- Undi95/MLewd-L2-13B-v1-3 (0.82)
|
|
- Sao10K/Stheno-Inverted-L2-13B (0.18)
|
|
|
|
!python ties_merge.py TheBloke/Llama-2-13B-fp16 ./MLewd-L2-13B-v1-7 --merge Undi95/MLewd-L2-13B-v1-3 --density 0.82 --merge Sao10K/Stheno-Inverted-L2-13B --density 0.18 --cuda
|
|
=> Result: MLewd-L2-13B-v1-7
|
|
|
|
|
|
Final touch (trying my best here) :
|
|
MLewd-L2-13B-v1-7 (0.77) + zarakiquemparte/PIPPA-ShareGPT-Subset-QLora-13b (LoRA 0.23)
|
|
=> MLewd-L2-13B-v1-7-TRY2
|
|
|
|
FINAL : MLewd-L2-13B-v1-7-TRY2 (0.82) + BluemoonRP (0.18)
|
|
=> MLewd-L2-13B-v1-8-3
|
|
|
|
RIP to all the version that got trashed.
|
|
```
|
|
|
|
<!-- description start -->
|
|
## Description
|
|
|
|
This repo contains fp16 files of MLewd-L2-13B, a trying-to-be lewd LLM model.
|
|
<!-- description end -->
|
|
<!-- description start -->
|
|
## Models used
|
|
- Undi95/ReMM (Base)
|
|
- Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged (Llama Chat Uncensored)
|
|
- KoboldAI/LLAMA2-13B-Holodeck-1
|
|
- Sao10K/Stheno-Inverted-L2-13B
|
|
## Loras used
|
|
- nRuaif/BluemoonRP-L2-13B-This-time-will-be-better/tree/main/lora-out-13b-final-BM/checkpoint-15/adapter_model
|
|
- zarakiquemparte/PIPPA-ShareGPT-Subset-QLora-13b
|
|
<!-- description end -->
|
|
<!-- prompt-template start -->
|
|
## Prompt template: Alpaca
|
|
|
|
```
|
|
Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
|
|
|
### Instruction:
|
|
{prompt}
|
|
|
|
### Response:
|
|
|
|
```
|
|
|
|
Special thanks to Sushi kek
|
|
# [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_Undi95__MLewd-L2-13B)
|
|
|
|
| Metric | Value |
|
|
|-----------------------|---------------------------|
|
|
| Avg. | 46.84 |
|
|
| ARC (25-shot) | 58.28 |
|
|
| HellaSwag (10-shot) | 82.32 |
|
|
| MMLU (5-shot) | 54.67 |
|
|
| TruthfulQA (0-shot) | 48.66 |
|
|
| Winogrande (5-shot) | 73.48 |
|
|
| GSM8K (5-shot) | 1.29 |
|
|
| DROP (3-shot) | 9.18 |
|