Qwen2-3x1.5B

Qwen2-3x1.5B is a Mixture of Experts (MoE) made with the following models using LazyMergekit:

• cognitivecomputations/dolphin-2.9.3-qwen2-1.5b
• Replete-AI/Qwen2-1.5b-Instruct-Replete-Adapted
• M4-ai/Hercules-5.0-Qwen2-1.5B

🧩 Configuration

gate_mode: hidden
architecture: qwen
dtype: bfloat16
experts_per_token: 2
base_model: M4-ai/Hercules-5.0-Qwen2-1.5B
experts:
  - source_model: cognitivecomputations/dolphin-2.9.3-qwen2-1.5b
    positive_prompts:
      - "chat"
      - "summarize"
      - "paraphrase"
      - "list"
      - "explain"
      - "define"
      - "analyze"
      - "rephrase"
      - "elaborate"
      - "You are an assistant"
      - "You are a summarization expert, condense lengthy documents, articles, and reports into brief, informative summaries."
      - "You are an academic tutor, provide detailed explanations of complex concepts in subjects like mathematics, physics, and biology."
      - "You are an educational consultant, offer guidance on study strategies and learning resources for various academic disciplines."
    negative_prompts:
      - "code"
      - "algorithm"
      - "programming"
  - source_model: Replete-AI/Qwen2-1.5b-Instruct-Replete-Adapted
    positive_prompts:
      - "programming language"
      - "JavaScript"
      - "Python programming language"
      - "Rust programming language"
      - "C++ programming language"
      - "GO programming language"
      - "Ruby programming language"
      - "Haskell programming language"
      - "SQL query language"
      - "CSS markup styling language"
      - "code"
      - "You are a software engineer, write efficient code in languages like Python, Java, and C++ to solve computational problems."
      - "You are a systems architect, design scalable and robust software systems, focusing on performance optimization and security."
      - "You are a function caller, able to execute functions and handle JSON data."
    negative_prompts:
      - "explain"
      - "describe"
      - "define"
  - source_model: M4-ai/Hercules-5.0-Qwen2-1.5B
    positive_prompts:
      - "characters"
      - "scene"
      - "roleplay"
      - "erotic roleplay"
      - "sexual fetish"
      - "NSFW"
      - "creative writing"
      - "storytelling"
      - "narration"
      - "narrative setting"
      - "narrative plot"
      - "narrative exposition"
      - "narrative theme"
      - "narrative climax"
      - "You are a content writer, create engaging articles, blog posts, and essays on topics ranging from technology to culture."
      - "You are a storyteller, capable of creating narratives and roleplaying scenarios."
      - "You are a creative writer, skilled in crafting stories and developing characters."
      - "You are a narrative designer, able to create engaging and immersive storylines."
      - "You are a language stylist, refine and polish written content to improve readability, coherence, and style."
      - "You are a knowledge synthesizer, provide concise summaries of historical events, scientific discoveries, and literary works."
    negative_prompts:
      - "code"
      - "algorithm"
      - "programming"
      - "summarize"
      - "paraphrase"
      - "list"
      - "explain"
      - "define"
      - "analyze"
      - "rephrase"
      - "elaborate"
      - "You are an assistant"
      - "explain"
      - "describe"
      - "define"
shared_experts:
  - source_model: M4-ai/Hercules-5.0-Qwen2-1.5B
    positive_prompts:
      - "conversationalist"
    negative_prompts:
      - "code"
      - "algorithm"
      - "programming"
    residual_scale: 0.1

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer, pipeline
import torch

model = "djuna/Qwen2-3x1.5B"

tokenizer = AutoTokenizer.from_pretrained(model)
generator = 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 = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = generator(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])