kuno-royale-v2-7b
An attempt to further strengthen the roleplaying prose of SanjiWatsuki/Kunoichi-DPO-v2-7B using eren23/ogno-monarch-jaskier-merge-7b-OH-PREF-DPO, a high-scorer for 7B models on the Open LLM Leaderboard.
Personal RP tests prove promising, and meaningless leaderboard metrics have improved vs SanjiWatsuki/Kunoichi-DPO-v2-7B.
Some GGUF quants available here.
Works well with Silly Tavern Noromaid template recommended by SanjiWatsuki for Kunoichi-7B: Context, Instruct
| 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 |
| core-3/kuno-royale-7B | 74.74 | 71.76 | 88.20 | 65.13 | 71.12 | 82.32 | 69.90 |
| 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 |
Original LazyMergekit Card:
kuno-royale-v2-7b is a merge of the following models using LazyMergekit:
🧩 Configuration
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
!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"])