bobchenyx/Qwen3-30B-A3B-Instruct-2507-MoBE-hf

Bobchenyx/MoBE/tree/Qwen3

For more usage instructions and details, please check my GitHub fork. https://github.com/Bobchenyx/MoBE/tree/Qwen3

MoBE: Mixture-of-Basis-Experts for Compressing
MoE-based LLMs

📘 Introduction

MoBE (Mixture-of-Basis-Experts) is a novel model compression technique designed for MoE LLMs developed by the AGI Center, Ant Group Research. It achieves efficient parameter reduction by factorizing each expert's weight matrix as:

$$\mathbf{W}=\mathbf{A}\mathbf{B},\text{where}\mathbf{B}=\sum _{i=1}^m{\alpha }_i{B}_i\backslash mathbf\{W\}\; =\; \backslash mathbf\{A\}\backslash mathbf\{B\},\; \backslash quad\; \backslash text\{where\}\; \backslash quad\; \backslash mathbf\{B\}\; =\; \backslash sum\_\{i=1\}^m\; \backslash alpha\_i\; B\_i$$

• $\mathbf{A}$: Expert-specific matrix
• $\mathbf{B}$: Linear combination of basis matrices across all experts, weighted by coefficients $\alpha_i$

The factorization is learned by minimizing the reconstruction error between the original and compressed weight matrices.

🔍 Key Results

MoBE significantly outperforms prior compression methods with minimal accuracy degradation:

• Reduces parameter count by 24%–30% in leading open-source models
• Incurs only 1%–2% absolute accuracy drop (≈2% relative)
• Demonstrated on Qwen3-235B, DeepSeek-V3 (671B), and Kimi-K2-Instruct (1T)

💡 MoBE Generate Example

from transformers import AutoTokenizer
from models.modeling_deepseek_v3_mobe import DeepseekV3MoBEForCausalLM 
from models.modeling_qwen3_mobe import Qwen3MoBEForCausalLM
from models.modeling_kimi_k2_mobe import KimiK2MoBEForCausalLM
import torch

model_name = "/root/DeepSeek-V3-0324-MoBE"
offload_folder = "./offload_dir"

tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.pad_token = tokenizer.eos_token

max_memory = {i: "120GiB" for i in range(8)}  
max_memory["cpu"] = "1200GiB"      

if 'Qwen' in model_name:
    model = Qwen3MoBEForCausalLM.from_pretrained(
            model_name,
            device_map="auto",
            offload_folder=offload_folder,
            offload_state_dict=True,
            torch_dtype=torch.bfloat16
        max_memory=max_memory
    )
elif 'DeepSeek' in model_name:
    model = DeepseekV3MoBEForCausalLM.from_pretrained(
            model_name,
            device_map="auto",
            offload_folder=offload_folder,
            offload_state_dict=True,
            torch_dtype=torch.bfloat16,
        max_memory=max_memory
    )
else:
        model = KimiK2MoBEForCausalLM.from_pretrained(
            model_name,
            device_map="auto",
            offload_folder=offload_folder,
            offload_state_dict=True,
            torch_dtype=torch.bfloat16,
        max_memory=max_memory
    )

input_text = "Artificial intelligence is"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")

with torch.no_grad():
outputs = model.generate(
        **inputs,
        max_new_tokens=128,
        do_sample=True,
        temperature=0.7,
        pad_token_id=tokenizer.eos_token_id
)

generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("Generated text:")
print(generated_text)

---

📚 Citation

If you find MoBE useful in your research or application, please consider citing our work:

@misc{chen2025mobemixtureofbasisexpertscompressingmoebased,
      title={MoBE: Mixture-of-Basis-Experts for Compressing MoE-based LLMs}, 
      author={Xiaodong Chen and Mingming Ha and Zhenzhong Lan and Jing Zhang and Jianguo Li},
      year={2025},
      eprint={2508.05257},
      archivePrefix={arXiv},
}