File size: 2,945 Bytes
670e3bf
 
 
 
 
 
 
 
 
 
ff99dd1
670e3bf
 
 
 
5542217
 
 
 
 
 
 
 
 
 
 
670e3bf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
tags:
- merge
- mergekit
- lazymergekit
- SanjiWatsuki/Kunoichi-DPO-v2-7B
- eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO
base_model:
- SanjiWatsuki/Kunoichi-DPO-v2-7B
- eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO
license: cc-by-nc-2.0
---

# kuno-royale-v2-7b

An attempt to further strengthen the roleplaying prose of [SanjiWatsuki/Kunoichi-DPO-v2-7B](https://huggingface.co./SanjiWatsuki/Kunoichi-DPO-v2-7B) using [eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO](https://huggingface.co./eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO), a high-scorer for 7B models on the [Open LLM Leaderboard](https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard).

Personal RP tests prove promising, and meaningless leaderboard metrics have improved vs [SanjiWatsuki/Kunoichi-DPO-v2-7B](https://huggingface.co./SanjiWatsuki/Kunoichi-DPO-v2-7B). 

|Model              | Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|-------------------|---------|-----|-----------|------|------------|------------|-------|
| eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO | 76.45 | 73.12 | 89.09 | 64.80 | 77.45 | 84.77 | 69.45 |
| **core-3/kuno-royale-v2-7b** | **74.80** | **72.01** | **88.15** | **65.07** | **71.10** | **82.24** | **70.20** |
| SanjiWatsuki/Kunoichi-DPO-v2-7B | 72.46 | 69.62 | 87.44 | 64.94 | 66.06 | 80.82 | 65.88 |
| SanjiWatsuki/Kunoichi-7B | 72.13 | 68.69 | 87.10 | 64.90 | 64.04 | 81.06 | 67.02 |

kuno-royale-v2-7b is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [SanjiWatsuki/Kunoichi-DPO-v2-7B](https://huggingface.co./SanjiWatsuki/Kunoichi-DPO-v2-7B)
* [eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO](https://huggingface.co./eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO)

## 🧩 Configuration

```yaml
slices:
  - sources:
      - model: SanjiWatsuki/Kunoichi-DPO-v2-7B
        layer_range: [0, 32]
      - model: eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO
        layer_range: [0, 32]
merge_method: slerp
base_model: SanjiWatsuki/Kunoichi-DPO-v2-7B
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5
dtype: bfloat16
```

## 💻 Usage

```python
!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "core-3/kuno-royale-v2-7b"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```