DaRuukLLM-Refresh-4x1B-v1

DaRuukLLM-Refresh-4x1B-v1 is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

• Trelis/Llama-3.2-1B-Instruct-MATH-3ep
• huihui-ai/Llama-3.2-1B-Instruct-abliterated
• passing2961/Ultron-Summarizer-1B
• unsloth/Llama-3.2-1B-Instruct

🧩 Configuration

base_model: unsloth/Llama-3.2-1B-Instruct  # Base model for self-attention and layer normalization
gate_mode: hidden  # Use hidden state representations for MoE gate parameters
dtype: bfloat16  # Output data type for the merged model

experts:
  - source_model: Trelis/Llama-3.2-1B-Instruct-MATH-3ep  # Expert for math-related tasks
    positive_prompts:
      - "Solve the following math problem:"
      - "Calculate the value of:"
      - "What is the result of:"

  - source_model: huihui-ai/Llama-3.2-1B-Instruct-abliterated  # Expert for uncensored queries
    positive_prompts:
      - "Explain the following controversial topic:"
      - "Discuss the implications of:"
      - "Provide an uncensored analysis of:"

  - source_model: passing2961/Ultron-Summarizer-1B  # Expert for summarization tasks
    positive_prompts:
      - "Summarize the following text:"
      - "Provide a concise summary of:"
      - "Generate a brief overview of:"

  - source_model: unsloth/Llama-3.2-1B-Instruct  # Base model also acts as the chat expert
    positive_prompts:
      - "How can I assist you today?"
      - "What would you like to discuss?"
      - "Let's have a conversation about:"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Xiaojian9992024/DaRuukLLM-Refresh-4x1B-v1"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
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"])