TinyQwex-4x620M-MoE

TinyQwex-4x620M-MoE is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

• Qwen/Qwen1.5-0.5B
• Qwen/Qwen1.5-0.5B
• Qwen/Qwen1.5-0.5B
• Qwen/Qwen1.5-0.5B

🌟 Buying me coffee is a direct way to show support for this project.

💻 Usage

!pip install -qU transformers bitsandbytes accelerate eniops

from transformers import AutoTokenizer
import transformers
import torch

model = "Isotonic/TinyQwex-4x620M-MoE"

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B")
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.bfloat16, "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"])

🧩 Configuration

experts:
  - source_model: Qwen/Qwen1.5-0.5B
    positive_prompts:
    - "reasoning"

  - source_model: Qwen/Qwen1.5-0.5B
    positive_prompts:
    - "program"

  - source_model: Qwen/Qwen1.5-0.5B
    positive_prompts:
    - "storytelling"

  - source_model: Qwen/Qwen1.5-0.5B
    positive_prompts:
    - "Instruction following assistant"