RepoInclusionAI (Ant Group)InclusionAI (Ant Group)published Aug 12, 2025seen 5d

inclusionAI/GroveMoE

Python

Open original ↗

Captured source

source ↗
published Aug 12, 2025seen 5dcaptured 14hhttp 200method plain

inclusionAI/GroveMoE

Language: Python

Stars: 25

Forks: 1

Open issues: 1

Created: 2025-08-12T02:48:59Z

Pushed: 2025-08-20T03:29:33Z

Default branch: master

Fork: no

Archived: no

README:

🤗 Models&nbsp&nbsp | &nbsp&nbsp 📑 Paper &nbsp&nbsp | &nbsp&nbsp 🔗 Github&nbsp&nbsp

Overview

GroveMoE is an open-source family of large language models developed by the AGI Center, Ant Group Research that introduces Grove MoE, a new sparse architecture using adjugate experts for dynamic computation allocation. With 33 B total parameters and 3.14–3.28 B active parameters per token, GroveMoE delivers state-of-the-art results across reasoning, mathematics, and code generation while keeping inference costs low.

---

Key Highlights

| Feature | Description | |---------|-------------| | Architecture | Novel adjugate experts grouped with ordinary experts; shared computation is executed once, then reused, cutting FLOPs. | | Sparse Activation | 33 B params total, only 3.14–3.28 B active per token. | | Training | Mid-training + SFT, up-cycled from Qwen3-30B-A3B-Base; preserves prior knowledge while adding new capabilities. | | Open | Weights, configs will be fully released under Apache 2.0 upon approval. |

---

Run GroveMoE

🤗 Transformers Quick Start

Transformers is a library of pretrained natural language processing for inference and training.

The following contains a code snippet illustrating how to use GroveMoE to generate content based on given inputs.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "inclusionAI/GroveMoE-Inst"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
**model_inputs,
max_new_tokens=16384
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()

content = tokenizer.decode(output_ids, skip_special_tokens=True)

print("content:", content)

🚀 SGLang Quick Start

SGLang is a fast serving framework for large language models and vision language models. SGLang could be used to launch a server with OpenAI-compatible API service.

1️⃣ Install Dependencies

First, install transformers:

cd src/transformers-4.51.3
pip install .

Then, install SGLang: 2. Install SGLang

cd src/sglang-0.4.6.post5
pip install .

2️⃣ Launch the Server

Run the following command to start SGLang:

python -m sglang.launch_server --model-path inclusionAI/GroveMoE-Inst --port 30000 --context-length 32768

3️⃣ Access the API

Once started, the OpenAI-compatible API will be available at http://localhost:30000/v1.

Test it with curl:

curl http://localhost:30000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "inclusionAI/GroveMoE-Inst",
"messages": [{"role": "user", "content": "Hello, SGLang!"}]
}'

Benchmark Results

---

Citation

If you find our work helpful, feel free to give us a cite.

@article{GroveMoE,
title = {GroveMoE: Towards Efficient and Superior MoE LLMs with Adjugate Experts},
author = {Wu, Haoyuan and Chen, Haoxing and Chen, Xiaodong and Zhou, Zhanchao and Chen, Tieyuan and Zhuang, Yihong and Lu, Guoshan and Zhao, Junbo and Liu, Lin and Huang, Zenan and Lan, Zhenzhong and Yu, Bei and Li, Jianguo},
journal = {arXiv preprint arXiv:2508.07785},
year = {2025}
}

Acknowledgements

We would like to thank the open-source community for their great contributions:

  • **Qwen3** for releasing high-quality foundation models.
  • **Transformers** for providing a versatile framework for training and inference.
  • **SGLang** for enabling efficient large-scale serving.

Notability

notability 3.0/10

New MoE repo, low stars.