Konstanta-7B

Konstanta-7B is a merge of the following models using LazyMergekit:

• maywell/PiVoT-0.1-Evil-a
• mlabonne/NeuralOmniBeagle-7B-v2

This is a test merge that is supposed to improve Kunoichi by merging it with new Beagle model and PiVoT Evil, which both show good performance. Even though the model's name is in Russian, it is not really capable of properly using it, as it was not the main goal of the model.

🧩 Configuration

merge_method: dare_ties
dtype: bfloat16
parameters:
  int8_mask: true
base_model: SanjiWatsuki/Kunoichi-DPO-v2-7B
models:
  - model: SanjiWatsuki/Kunoichi-DPO-v2-7B
  - model: maywell/PiVoT-0.1-Evil-a
    parameters:
      density: 0.65
      weight: 0.15
  - model: mlabonne/NeuralOmniBeagle-7B-v2
    parameters:
      density: 0.85
      weight: 0.45

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Inv/Konstanta-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"])

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	73.54
AI2 Reasoning Challenge (25-Shot)	70.05
HellaSwag (10-Shot)	87.50
MMLU (5-Shot)	65.06
TruthfulQA (0-shot)	65.43
Winogrande (5-shot)	82.16
GSM8k (5-shot)	71.04