silencer107/bobik7b

RolePlayLake-7B

RolePlayLake-7B is a merge of the following models :

• SanjiWatsuki/Silicon-Maid-7B
• senseable/WestLake-7B-v2

In my current testing RolePlayLake is Better than Silicon_Maid in RP and More Uncensored Than WestLake

I would try to only merge Uncensored Models with Baising towards Chat rather than Instruct

🧩 Configuration

slices:
  - sources:
      - model: SanjiWatsuki/Silicon-Maid-7B
        layer_range: [0, 32]
      - model: senseable/WestLake-7B-v2
        layer_range: [0, 32]
merge_method: slerp
base_model: senseable/WestLake-7B-v2
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 = "fhai50032/RolePlayLake-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"])

Why I Merged WestLake and Silicon Maid

Merged WestLake and Silicon Maid for a unique blend:

1. EQ-Bench Dominance: WestLake's 79.75 EQ-Bench score. (Maybe Contaminated)
2. Charm and Role-Play: Silicon's explicit charm and WestLake's role-play prowess.
3. Config Synergy: Supports lots of prompt format out of the gate and has a very nice synergy

Result: RolePlayLake-7B, a linguistic fusion with EQ-Bench supremacy and captivating role-play potential.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	72.54
AI2 Reasoning Challenge (25-Shot)	70.56
HellaSwag (10-Shot)	87.42
MMLU (5-Shot)	64.55
TruthfulQA (0-shot)	64.38
Winogrande (5-shot)	83.27
GSM8k (5-shot)	65.05