Views
No views yet
1Due to hardware limitation, some merge was done in 2 part.
2
3Last mix :
4- ReMM (Base) (0.57)
5- Doctor-Shotgun/llama-2-13b-chat-limarp-v2-merged (Llama Chat Uncensored) (0.35)
6- KoboldAI/LLAMA2-13B-Holodeck-1 (0.08)
7
8Part 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
9
10Part 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
11
12(MLewd-L2-13B-v1-2 got disqualified)
13
14- Applying LoRA: nRuaif/Kimiko-v2-13B at (0.24) weight on MLewd-L2-13B-v1-1
15=> Result: MLewd-L2-13B-v1-3
16
17================== ERP RANKING TEST ===========================
18
1919.42 | MLewd-L2-13B-v1-3.q5_K_M.gguf (-> Best)
2019.25 | MLewd-L2-13B-v1-1.q5_K_M.gguf
2118.25 | MLewd-L2-13B-v1-2.q5 K M.gguf
22
23================== RETRY ===========================
24
25Mix:
26- Undi95/MLewd-L2-13B-v1-3 (0.82)
27- Sao10K/Stheno-Inverted-L2-13B (0.18)
28
29!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
30=> Result: MLewd-L2-13B-v1-7
31
32
33Final touch (trying my best here) :
34MLewd-L2-13B-v1-7 (0.77) + zarakiquemparte/PIPPA-ShareGPT-Subset-QLora-13b (LoRA 0.23)
35=> MLewd-L2-13B-v1-7-TRY2
36
37FINAL : MLewd-L2-13B-v1-7-TRY2 (0.82) + BluemoonRP (0.18)
38=> MLewd-L2-13B-v1-8-3
39
40RIP to all the version that got trashed.Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{prompt}
### Response:
| 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 |