LoneStriker/Wiedervereinigung-7b-dpo-4.0bpw-h6-exl2

Wiedervereinigung-7b-dpo

This is a dpo aligned merge of multiple german models scoring 7.1 on the mt-bench-de average. It is a merge of the best german 7B models with 7b parameters as a dare_ties merge. Since the original models based on mistral - three of them on the brilliant german LeoLM/leo-mistral-hessianai-7b - they are reunited in this merged model. Therefore the name, no nationalist ideas involved. To improve result quality they are dpo-trained with a german translation of intel-orca-dpo using our german fork of LLaMA-Factory.

mt-bench-de

Is the merged model good? Well, of course. But it is even better with the help of some dpo tuning.

{
    "first_turn": 7.3,
    "second_turn": 6.925,
    "categories": {
        "writing": 8.425,
        "roleplay": 8.6,
        "reasoning": 5.4,
        "math": 4.35,
        "coding": 4.3,
        "extraction": 7.975,
        "stem": 8.5,
        "humanities": 9.35
    },
    "average": 7.1125
}

Wiedervereinigung-7b itself is a LazyMergekit merge of:

• DiscoResearch/DiscoLM_German_7b_v1
• DRXD1000/Phoenix
• VAGOsolutions/SauerkrautLM-7b-v1-mistral
• malteos/hermeo-7b

All the actual heavylifting has been done by the creators of these models.

🧩 Configuration

models:
  - model: LeoLM/leo-mistral-hessianai-7b
    # No parameters necessary for base model
  - model: DiscoResearch/DiscoLM_German_7b_v1
    parameters:
      density: 0.6
      weight: 0.25
  - model: DRXD1000/Phoenix
    parameters:
      density: 0.6
      weight: 0.25
  - model: VAGOsolutions/SauerkrautLM-7b-v1-mistral
    parameters:
      density: 0.6
      weight: 0.25
  - model: malteos/hermeo-7b
    parameters:
      density: 0.6
      weight: 0.25
merge_method: dare_ties
base_model: LeoLM/leo-mistral-hessianai-7b
parameters:
  int8_mask: true
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mayflowergmbh/Wiedervereinigung-7b-dpo"
messages = [{"role": "user", "content": "Was ist ein deutsches 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"])