MiniMax-VL-01

1. Introduction

We are delighted to introduce our MiniMax-VL-01 model. It adopts the “ViT-MLP-LLM” framework, which is a commonly used technique in the field of multimodal large language models. The model is initialized and trained with three key parts: a 303-million-parameter Vision Transformer (ViT) for visual encoding, a randomly initialized two-layer MLP projector for image adaptation, and the MiniMax-Text-01 as the base LLM. MiniMax-VL-01 has a notable dynamic resolution feature. Input images are resized per a pre-set grid, with resolutions from 336×336 to 2016×2016, keeping a 336×336 thumbnail. The resized images are split into non-overlapping patches of the same size. These patches and the thumbnail are encoded separately and then combined for a full image representation. The training data for MiniMax-VL-01 consists of caption, description, and instruction data. The Vision Transformer (ViT) is trained on 694 million image-caption pairs from scratch. Across four distinct stages of the training pipeline, a total of 512 billion tokens are processed, leveraging this vast amount of data to endow the model with strong capabilities. Finally, MiniMax-VL-01 has reached top-level performance on multimodal leaderboards, demonstrating its edge and dependability in complex multimodal tasks.

2. Evaluation

Tasks	GPT-4o (11-20)	Claude-3.5-Sonnet (10-22)	Gemini-1.5-Pro (002)	Gemini-2.0-Flash (exp)	Qwen2-VL-72B-Inst.	InternVL2.5-78B	LLama-3.2-90B	MiniMax-VL-01
Knowledge
MMMU^*	63.5	72.0	68.4	70.6	64.5	66.5	62.1	68.5
MMMU-Pro^*	54.5	54.7	50.9	57.0	43.2	47.3	36.0	52.7
Visual Q&A
ChartQA^*_relaxed	88.1	90.8	88.7	88.3	91.2	91.5	85.5	91.7
DocVQA^*	91.1	94.2	91.5	92.9	97.1	96.1	90.1	96.4
OCRBench	806	790	800	846	856	847	805	865
Mathematics & Sciences
AI2D^*	83.1	82.0	80.9	85.1	84.4	86.8	78.9	83.3
MathVista^*	62.1	65.4	70.6	73.1	69.6	68.4	57.3	68.6
OlympiadBench_full	25.2	28.4	32.1	46.1	21.9	25.1	19.3	24.2
Long Context
M-LongDoc_acc	41.4	31.4	26.2	31.4	11.6	19.7	13.9	32.5
Comprehensive
MEGA-Bench_macro	49.4	51.4	45.9	53.9	46.8	45.3	19.9	47.4
User Experience
In-house Benchmark	62.3	47.0	49.2	72.1	40.6	34.8	13.6	56.6

^* Evaluated following a 0-shot CoT setting.

3. Quickstart

Here we provide a simple example of loading the tokenizer and model to generate content.

from transformers import AutoModelForCausalLM, AutoProcessor, AutoConfig, QuantoConfig, GenerationConfig
import torch
import json
import os
from PIL import Image

# load hf config
hf_config = AutoConfig.from_pretrained("MiniMaxAI/MiniMax-VL-01", trust_remote_code=True)

# quantization config, int8 is recommended
quantization_config =  QuantoConfig(
            weights="int8",
            modules_to_not_convert=[
                "vision_tower",
                "image_newline",
                "multi_modal_projector",
                "lm_head",
                "embed_tokens",
            ] + [f"model.layers.{i}.coefficient" for i in range(hf_config.text_config.num_hidden_layers)]
            + [f"model.layers.{i}.block_sparse_moe.gate" for i in range(hf_config.text_config.num_hidden_layers)]
        )

# set device map
model_safetensors_index_path = os.path.join("MiniMax-VL-01", "model.safetensors.index.json")
with open(model_safetensors_index_path, "r") as f:
    model_safetensors_index = json.load(f)
weight_map = model_safetensors_index['weight_map']
vision_map = {}
for key, value in weight_map.items():
    if 'vision_tower' in key or 'image_newline' in key or 'multi_modal_projector' in key:
        new_key = key.replace('.weight','').replace('.bias','')
        if new_key not in vision_map:
            vision_map[new_key] = value
# assume 8 GPUs
world_size = 8
device_map = {
    'language_model.model.embed_tokens': 'cuda:0',
    'language_model.model.norm': f'cuda:{world_size - 1}',
    'language_model.lm_head': f'cuda:{world_size - 1}'
}
for key, value in vision_map.items():
    device_map[key] = f'cuda:0'
device_map['vision_tower.vision_model.post_layernorm'] = f'cuda:0'
layers_per_device = hf_config.text_config.num_hidden_layers // world_size
for i in range(world_size):
    for j in range(layers_per_device):
        device_map[f'language_model.model.layers.{i * layers_per_device + j}'] = f'cuda:{i}'

# load processor
processor = AutoProcessor.from_pretrained("MiniMaxAI/MiniMax-VL-01", trust_remote_code=True)
messages = [
    {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant created by MiniMax based on MiniMax-VL-01 model."}]},
    {"role": "user", "content": [{"type": "image", "image": "placeholder"},{"type": "text", "text": "Describe this image."}]},
]
prompt = processor.tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
raw_image = Image.open("figures/image.jpg")
# tokenize and move to device
model_inputs = processor(images=[raw_image], text=prompt, return_tensors='pt').to('cuda').to(torch.bfloat16)

# load bfloat16 model, move to device, and apply quantization
quantized_model = AutoModelForCausalLM.from_pretrained(
    "MiniMaxAI/MiniMax-VL-01",
    torch_dtype="bfloat16",
    device_map=device_map,
    quantization_config=quantization_config,
    trust_remote_code=True,
    offload_buffers=True,
)
generation_config = GenerationConfig(
    max_new_tokens=100,
    eos_token_id=200020,
    use_cache=True,
)

# generate response
generated_ids = quantized_model.generate(**model_inputs, generation_config=generation_config)
print(f"generated_ids: {generated_ids}")
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = processor.tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

4. Citation

@misc{minimax2025minimax01scalingfoundationmodels,
      title={MiniMax-01: Scaling Foundation Models with Lightning Attention}, 
      author={MiniMax and Aonian Li and Bangwei Gong and Bo Yang and Boji Shan and Chang Liu and Cheng Zhu and Chunhao Zhang and Congchao Guo and Da Chen and Dong Li and Enwei Jiao and Gengxin Li and Guojun Zhang and Haohai Sun and Houze Dong and Jiadai Zhu and Jiaqi Zhuang and Jiayuan Song and Jin Zhu and Jingtao Han and Jingyang Li and Junbin Xie and Junhao Xu and Junjie Yan and Kaishun Zhang and Kecheng Xiao and Kexi Kang and Le Han and Leyang Wang and Lianfei Yu and Liheng Feng and Lin Zheng and Linbo Chai and Long Xing and Meizhi Ju and Mingyuan Chi and Mozhi Zhang and Peikai Huang and Pengcheng Niu and Pengfei Li and Pengyu Zhao and Qi Yang and Qidi Xu and Qiexiang Wang and Qin Wang and Qiuhui Li and Ruitao Leng and Shengmin Shi and Shuqi Yu and Sichen Li and Songquan Zhu and Tao Huang and Tianrun Liang and Weigao Sun and Weixuan Sun and Weiyu Cheng and Wenkai Li and Xiangjun Song and Xiao Su and Xiaodong Han and Xinjie Zhang and Xinzhu Hou and Xu Min and Xun Zou and Xuyang Shen and Yan Gong and Yingjie Zhu and Yipeng Zhou and Yiran Zhong and Yongyi Hu and Yuanxiang Fan and Yue Yu and Yufeng Yang and Yuhao Li and Yunan Huang and Yunji Li and Yunpeng Huang and Yunzhi Xu and Yuxin Mao and Zehan Li and Zekang Li and Zewei Tao and Zewen Ying and Zhaoyang Cong and Zhen Qin and Zhenhua Fan and Zhihang Yu and Zhuo Jiang and Zijia Wu},
      year={2025},
      eprint={2501.08313},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2501.08313}, 
}

5. Chatbot & API

For general use and evaluation, we provide a Chatbot with online search capabilities and the online API for developers.