Hugging Face's logo

inclusionAI
/
Ming-Lite-Omni

Any-to-Any
Transformers
ONNX
Diffusers
Safetensors
bailingmm
text-generation
custom_code
Model card Files
xet
 Community
6
Ming-Lite-Omni / README.md
qingpei's picture
qingpei
Update README.md
ecd74fd verified
preview code
|
Raw
History Blame Contribute Delete
25.3 kB
---
base_model:
- inclusionAI/Ling-lite
license: mit
pipeline_tag: any-to-any
library_name: transformers
new_version: inclusionAI/Ming-flash-omni-Preview
---
# Ming-Lite-Omni
<p align="center">
 <img src="./figures/ant-bailing.png" width="100"/>
<p>
<p align="center">📑 <a href="https://arxiv.org/abs/2506.09344">Technical Report</a>｜📖<a href="https://lucaria-academy.github.io/Ming-Omni/">Project Page</a> ｜🤗 <a href="https://huggingface.co/inclusionAI/Ming-Lite-Omni">Hugging Face</a>｜ 🤖 <a href="https://www.modelscope.cn/models/inclusionAI/Ming-Lite-Omni">ModelScope</a>
## Introduction
Ming-lite-omni, a light version of Ming-omni, which is derived from [Ling-lite](https://github.com/inclusionAI/Ling) and features 2.8 billion activated parameter. Ming-lite-omni is a unified multimodal model capable of processing images, text, audio, and video, while demonstrating strong proficiency in both speech and image generation. Ming-lite-omni employs dedicated encoders to extract tokens from different modalities, which are then processed by Ling, an MoE architecture equipped with newly proposed modality-specific routers. This design enables a single model to efficiently process and fuse multimodal inputs within a unified framework, thereby facilitating diverse tasks without requiring separate models, task-specific fine-tuning, or structural redesign. Importantly, Ming-lite-omni extends beyond conventional multimodal models by supporting audio and image generation. This is achieved through the integration of an advanced audio decoder for natural-sounding speech and Ming-Lite-Uni for high-quality image generation, which also allow the model to engage in context-aware chatting, perform text-to-speech conversion, and conduct versatile image editing. Our experimental results showcase Ming-lite-omni offers a powerful solution for unified perception and generation across all modalities.
Notably, Ming-lite-omni is the first open-source model we are aware of to match GPT-4o in modality support, and we release all code and model weights to encourage further research and development in the community.
<p align="center">
 <img src="./figures/ming.png" width="800"/>
<p>
## 📌 Updates
* [2025.06.12] 🔥 Our [Technical Report](https://arxiv.org/abs/2506.09344) is in public on arxiv.
* [2025.05.28] 🔥 The official version of Ming-lite-omni is released, with better performance and image generation support.
* [2025.05.04] 🔥 We release the test version of Ming-lite-omni：[Ming-lite-omni-Preview](https://github.com/inclusionAI/Ming/tree/Ming-Lite-Omni-Preview).
## Key Features
- **Unified Omni-Modality Perception**: Ming-lite-omni, built on [Ling](https://github.com/inclusionAI/Ling), an MoE architecture LLM, resolves task conflicts and ensures coherent integration of tokens from different modalities through modality-specific routers.
- **Unified Perception and Generation**: Ming-lite-omni achieves unified understanding and generation, enabling the model to interpret multimodal instructions and user intent during generation, which helps enhance generation quality and improves usability across multiple tasks.
- **Innovative Generation Capabilities**: Ming-lite-omni can perceive all modalities and generate high-quality text, real-time speech, and vivid images simultaneously, delivering exceptional cross-modal performance across diverse tasks including image perception, audio-visual interaction, and image generation.
## Evaluation
Ming-lite-omni delivers exceptional cross-modal performance, as validated across image perception, audio-visual interaction, and image generation tasks. Specifically, in the image perception task, Ming-lite-omni attained performance comparable to that of Qwen2.5-VL-7B by activating only 2.8B parameters. It delivers superior performance in end-to-end speech understanding and instruction following, surpassing Qwen2.5-Omni and Kimi-Audio. It also supports native-resolution image generation, editing, and style transfer, achieving a GenEval score of 0.64, outperforming mainstream models such as SDXL. In terms of FID, Ming-lite-omni reaches 4.85, setting a new SOTA across existing methods.
<p align="center">
 <img src="./figures/performance.png" width="800"/>
<p>
### Image benchmark
<div align="center">
| Benchmarks | Ming-lite-omni | Qwen2.5-VL-7B-Instruct | InternVL2.5-8B-MPO |
|:------------------|:--------------:|:----------------------------:|:------------------:|
| AI2D | 83.1 | 84.4 | <b>84.5</b> |
| HallusionBench | <b>55.0</b> | 55.8 | 51.7 |
| MMBench_TEST_V11 | 80.8 | <b>82.8</b> | 82.0 |
| MMMU | 56.3 | <b>56.6</b> | 54.8 |
| MMStar | 64.7 | 65.3 | <b>65.2</b> |
| MMVet | 71.3 | 71.6 | 68.1 |
| MathVista | <b>71.6</b> | 68.1 | 67.9 |
| OCRBench | <b>88.4</b> | 87.8 | 88.2 |
| Average | 71.4 | <b>71.5</b> | 70.3 |
</div>
#### Encyclopedia Benchmarks
<div align="center">
| Object Recognition | Ming-lite-omni | Qwen2.5-VL-7B-Instruct |
|:---------------------|:--------------:|:------------------------:|
| Plants | **54.96** | 47.8 |
| Animals | **56.7** | 50.85 |
| Vehicles | 41.91 | **42.29** |
| Food & Ingredients | **62.28** | 54.09 |
| Dishes | **44.3** | 39.07 |
| General | 91.08 | **92.42** |
| Average | **58.54** | 54.43 |
</div>
### Video benchmark
<div align="center">
| Benchmarks | Ming-lite-omni | Qwen2.5VL-7B-Instruct |
|:------------------------|:--------------:|:---------------------:|
| VideoMME | 67.0 | <b>67.3</b> |
| MVBench | 67.7 | <b>67.4</b> |
| Video-MMMU | 46.3 | <b>47.4</b> |
| LongVideoBench | 56.6 | 54.7 |
| Average | <b>59.4</b> | 59.2 |
</div>
Note: All models are evaluated based on 128 uniformly sampled frames.
### Audio benchmark
#### SpeechQA
<div align="center">
| Model | Average | AlpacaEval | CommonEval | SD-QA | MMSU | OpenBookQA | IFEval | AdvBench |
|:-----------------|:-------------:|:-----------:|:-----------:|:------------:|:------------:|:------------:|:------------:|:-------------:|
| Qwen2-Audio-chat | 3.545 | 3.69 | 3.40 | 35.35 | 35.43 | 49.01 | 22.57 | 98.85 |
| Baichuan-Audio | 3.695 | 4.00 | 3.39 | 49.64 | 48.80 | 63.30 | 41.32 | 86.73 |
| GLM-4-Voice | 3.77 | 4.06 | 3.48 | 43.31 | 40.11 | 52.97 | 24.91 | 88.08 |
| Kimi-Audio | 4.215 | 4.46 | 3.97 | <b>63.12</b> | <b>62.17</b> | <b>83.52</b> | <b>61.10</b> | <b>100.00</b> |
| Qwen2.5-Omni | 4.21 | 4.49 | 3.93 | 55.71 | 61.32 | 81.10 | 52.87 | 99.42 |
| Ming-lite-omni | <b>4.34</b> | <b>4.63</b> | <b>4.06</b> | 58.84 | 47.53 | 61.98 | 58.36 | 99.04 |
</div>
#### ASR
<div align="center">
| Model | aishell1 | aishell2_android | aishell2_ios | cv15_zh | fleurs_zh | wenetspeech_meeting | wenetspeech_net | librispeech_test_clean | librispeech_test_other | multilingual_librispeech | cv15_en | fleurs_en | voxpopuli_v1.0_en |
|:--------------:|:--------:|:----------------:|:------------:|:--------:|:---------:|:-------------------:|:---------------:|:----------------------:|:----------------------:|:------------------------:|:--------:|:---------:|:--------------------:|
| Ming-lite-omni | 1.47 | **2.55** | **2.52** | 6.31 | 2.96 | 5.95 | 5.46 | 1.44 | 2.80 | **4.15** | **6.89** | **3.39** | **5.80** |
| Qwen2.-Omni | 1.18 | 2.75 | 2.63 | **5.20** | 3.00 | **5.90** | 7.70 | 1.80 | 3.40 | 7.56 | 7.60 | 4.10 | **5.80** |
| Qwen2-Audio | 1.53 | 2.92 | 2.92 | 6.90 | 7.50 | 7.16 | 8.42 | 1.60 | 3.60 | 5.40 | 8.60 | 6.90 | 6.84 |
| Kimi-Audio | **0.60** | 2.64 | 2.56 | 7.21 | **2.69** | 6.28 | **5.37** | **1.28** | **2.42** | 5.88 | 10.31 | 4.44 | 7.97 |
</div>
### Information-Seeking Benchmark
<div align="center">
| Model | InfoSeek_H-mean | InfoSeek_unseen_question | InfoSeek_unseen_entity |
|:---------------|:---------------:|:------------------------:|:----------------------:|
| GPT-4o | <b>36.05</b> | - | - |
| PaLI-X | 22.06 | 23.5 | 20.8 |
| Qwen2.5-vl-32B | 19.35 | 20.55 | 18.28 |
| Ming-lite-omni | 27.7 | **30.4** | **25.4** |
</div>
### OCR
<div align="center">
| Model | Ming-lite-omni | Qwen2.5-VL-7B-Instruct |
|:-------------------|:--------------:|:-----------------------:|
| ChartQA_TEST | 85.1 | <b>87.3</b> |
| DocVQA_TEST | 93 | <b>95.7</b> |
| OCRBenchV2_en/zh | 53.3/52 | <b>56.3/57.2</b> |
| OmniDocBench↓ | 34/<b>34.4</b> | <b>30.8</b>/39.8 |
| TextVQA_VAL | 82.8 | <b>84.9</b> |
</div>
### GUI
<div align="center">
| Model | Ming-lite-omni | InternVL3 8B | Qwen2.5-VL-7B-Instruct |
|:---------------------------|:--------------:|:------------:|:----------------------:|
| ScreenSpot | <b>82.1</b> | 79.5 | 78.9* |
| ScreenSpot-V2 | <b>84.1</b> | 81.4 | - |
| AITZ(EM) | <b>66.6</b> | - | 57.6* |
</div>
Note: * denotes the reproduced results.
### Unified Generation Benchmark
<div align="center">
| Model | single_object | two_object | counting | colors | position | color_attr | GENEVAL | DPGBench | FID↓ |
|:---------------|:-------------:|:----------:|:----------:|:--------:|:--------:|:----------:|:--------:|:---------:|:-------------:|
| Ming-lite-omni | **0.9875** | **0.7727** | **0.6812** | 0.7872 | 0.31 | 0.29 | **0.64** | 81.72 | **4.85** |
| Metaquery-XL | - | - | - | - | - | - | 0.61 | **82.05** | 6.02 |
| SDv2.1 | 0.98 | 0.51 | 0.44 | **0.85** | 0.07 | 0.17 | 0.50 | 68.09 | 26.96 |
| Emu3-Gen | 0.98 | 0.71 | 0.34 | 0.81 | 0.17 | 0.21 | 0.54 | 80.60 | - |
| SDXL | 0.98 | 0.74 | 0.39 | **0.85** | 0.15 | 0.23 | 0.55 | 74.65 | 8.76 |
| Janus | 0.97 | 0.68 | 0.30 | 0.84 | **0.46** | **0.42** | 0.61 | 79.68 | 10.10 |
| JanusFlow | - | - | - | - | - | - | 0.63 | 80.09 | 9.51 |
</div>
Please refer to our technical report for more comprehensive evaluation results.
## Model Downloads
You can download the model from both Huggingface and ModelScope.
<div align="center">
| **Model** | **Input modality** | **Oput modality** | **Download** |
|:---------------| :---------------------: | :---------------: |:----------------------------------------------------------------------------------------------------------------------------------------------------:|
| Ming-Lite-Omni | Image,text,viedio,audio | Image,text,audio | [🤗 HuggingFace](https://huggingface.co/inclusionAI/Ming-Lite-Omni) <br>[🤖 ModelScope](https://www.modelscope.cn/models/inclusionAI/Ming-Lite-Omni) |
</div>
If you're in mainland China, we strongly recommend you to download our model from 🤖 <a href="https://www.modelscope.cn/models/inclusionAI/Ming-Lite-Omni">ModelScope</a>.
## Use Cases
Additional demonstration cases are available on our project [page](https://lucaria-academy.github.io/Ming-Omni/).
## Example Usage
Please download our model following [Model Downloads](#model-downloads), then you can refer to the following codes to run Ming-lite-omni model.
Python environment dependency installation.
```shell
pip install -r requirements.txt
pip install data/matcha_tts-0.0.5.1-cp38-cp38-linux_x86_64.whl
pip install diffusers==0.33.0
pip install nvidia-cublas-cu12==12.4.5.8 # for H20
```
Note: We test following examples on hardware of NVIDIA H800-80GB with CUDA 12.2. Loading inclusionAI/Ming-Lite-Omni in bfloat16 takes about 40890MB memory.
```python
import os
import torch
from transformers import AutoProcessor, GenerationConfig
from modeling_bailingmm import BailingMMNativeForConditionalGeneration
# build model
model = BailingMMNativeForConditionalGeneration.from_pretrained(
 "inclusionAI/Ming-Lite-Omni",
 torch_dtype=torch.bfloat16,
 low_cpu_mem_usage=True
).to("cuda")
assets_path = YOUR_ASSETS_PATH
# build processor
processor = AutoProcessor.from_pretrained("inclusionAI/Ming-Lite-Omni", trust_remote_code=True)
```
```python
# qa
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "text", "text": "请详细介绍鹦鹉的生活习性。"}
 ],
 },
]
# Output:
# 鹦鹉是一种非常聪明和社交性强的鸟类，它们的生活习性非常丰富和有趣。以下是一些关于鹦鹉生活习性的详细介绍：
# ### 1. **栖息地**
# 鹦鹉主要分布在热带和亚热带地区，包括非洲、亚洲、澳大利亚和南美洲。它们通常生活在森林、草原、沙漠和城市环境中。不同种类的鹦鹉对栖息地的要求有所不同，但大多数鹦鹉喜欢有丰富植被和水源的地方。
# ### 2. **饮食**
# 鹦鹉是杂食性动物，它们的饮食非常多样化。它们的食物包括种子、坚果、水果、蔬菜、花蜜和昆虫。鹦鹉的喙非常强壮，能够轻松地打开坚硬的果壳和坚果。一些鹦鹉还会吃泥土或沙子，以帮助消化和补充矿物质。
# ......
```
```python
# image qa
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "image", "image": os.path.join(assets_path, "flowers.jpg")},
 {"type": "text", "text": "What kind of flower is this?"},
 ],
 },
]
# Output:
# The flowers in this image are forget-me-nots. These delicate blooms are known for their small, five-petaled flowers that come in various shades of blue, pink, and white.
```
To enable thinking before response, adding the following system prompt before your question:
```python
cot_prompt = "SYSTEM: You are a helpful assistant. When the user asks a question, your response must include two parts: first, the reasoning process enclosed in <thinking>...</thinking> tags, then the final answer enclosed in <answer>...</answer> tags. The critical answer or key result should be placed within \\boxed{}.\n"
# And your input message should be like this:
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "image", "image": os.path.join(assets_path, "reasoning.png")},
 {"type": "text", "text": cot_prompt + "In the rectangle $A B C D$ pictured, $M_{1}$ is the midpoint of $D C, M_{2}$ the midpoint of $A M_{1}, M_{3}$ the midpoint of $B M_{2}$ and $M_{4}$ the midpoint of $C M_{3}$. Determine the ratio of the area of the quadrilateral $M_{1} M_{2} M_{3} M_{4}$ to the area of the rectangle $A B C D$.\nChoices:\n(A) $\frac{7}{16}$\n(B) $\frac{3}{16}$\n(C) $\frac{7}{32}$\n(D) $\frac{9}{32}$\n(E) $\frac{1}{5}$"},
 ],
 },
]
# Output:
# \<think\>\nOkay, so I have this problem about a rectangle ABCD ... (thinking process omitted) ... So, the correct answer is C.\n\</think\>\n\<answer\>\\boxed{C}\</answer\>\n\n
```
```python
# video qa
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "video", "video": os.path.join(assets_path, "yoga.mp4")},
 {"type": "text", "text": "What is the woman doing?"},
 ],
 },
]
# Output:
# The image shows a woman performing a yoga pose on a rooftop. She's in a dynamic yoga pose, with her arms and legs extended in various positions.
```
```python
# multi-turn chat
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "text", "text": "中国的首都是哪里？"},
 ],
 },
 {
 "role": "ASSISTANT",
 "content": [
 {"type": "text", "text": "北京"},
 ],
 },
 {
 "role": "HUMAN",
 "content": [
 {"type": "text", "text": "它的占地面积是多少？有多少常住人口？"},
 ],
 },
]
# Output:
# 北京市的总面积约为16,410.54平方公里，常住人口约为21,542,000人。
```
```python
# Preparation for inference
text = processor.apply_chat_template(messages, add_generation_prompt=True)
image_inputs, video_inputs, audio_inputs = processor.process_vision_info(messages)
inputs = processor(
 text=[text],
 images=image_inputs,
 videos=video_inputs,
 audios=audio_inputs,
 return_tensors="pt",
)
inputs = inputs.to(model.device)
for k in inputs.keys():
 if k == "pixel_values" or k == "pixel_values_videos" or k == "audio_feats":
 inputs[k] = inputs[k].to(dtype=torch.bfloat16)
# call generate
generation_config = GenerationConfig.from_dict({'no_repeat_ngram_size': 10})
generated_ids = model.generate(
 **inputs,
 max_new_tokens=512,
 use_cache=True,
 eos_token_id=processor.gen_terminator,
 generation_config=generation_config,
)
generated_ids_trimmed = [
 out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
 ]
output_text = processor.batch_decode(
 generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
print(output_text)
```
### Audio tasks
```python
# ASR
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "text", "text": "Please recognize the language of this speech and transcribe it. Format: oral."},
 {"type": "audio", "audio": 'data/wavs/BAC009S0915W0292.wav'},
 ],
 },
]
# we use whisper encoder for ASR task, so need modify code above
inputs = processor(
 text=[text],
 images=image_inputs,
 videos=video_inputs,
 audios=audio_inputs,
 return_tensors="pt",
 audio_kwargs={'use_whisper_encoder': True}
)
outputs = model.generate(
 **inputs,
 max_new_tokens=512,
 use_cache=True,
 eos_token_id=processor.gen_terminator,
 generation_config=generation_config,
 use_whisper_encoder=True
)
```
```python
# speech2speech
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "audio", "audio": 'data/wavs/speechQA_sample.wav'},
 ],
 },
]
generation_config = GenerationConfig.from_dict({
 'output_hidden_states': True,
 'return_dict_in_generate': True,
 'no_repeat_ngram_size': 10}
)
outputs = model.generate(
 **inputs,
 max_new_tokens=512,
 use_cache=True,
 eos_token_id=processor.gen_terminator,
 generation_config=generation_config,
 use_whisper_encoder=False
)
generated_ids = outputs.sequences
generated_ids_trimmed = [
 out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
# speechQA result
output_text = processor.batch_decode(
 generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
# for TTS
from modeling_bailing_talker import AudioDetokenizer
model_name_or_path = model.config._name_or_path
audio_detokenizer = AudioDetokenizer(
 f'{model_name_or_path}/talker/audio_detokenizer.yaml',
 flow_model_path=f'{model_name_or_path}/talker/flow.pt',
 hifigan_model_path=f'{model_name_or_path}/talker/hift.pt'
)
spk_input = torch.load('data/spks/luna.pt')
thinker_reply_part = outputs.hidden_states[0][0] + outputs.hidden_states[0][-1]
# Setting thinker_reply_part to None allows the talker to operate as a standalone TTS model, independent of the language model.
audio_tokens = model.talker.omni_audio_generation(
 output_text,
thinker_reply_part=thinker_reply_part, **spk_input)
waveform = audio_detokenizer.token2wav(audio_tokens, save_path='out.wav', **spk_input)
```
```python
# zero-shot TTS
from modeling_bailing_talker import AudioDetokenizer
from audio_detokenizer.cli.frontend import TTSFrontEnd
from hyperpyyaml import load_hyperpyyaml
model_name_or_path = model.config._name_or_path
audio_detokenizer = AudioDetokenizer(
 f'{model_name_or_path}/talker/audio_detokenizer.yaml',
 flow_model_path=f'{model_name_or_path}/talker/flow.pt',
 hifigan_model_path=f'{model_name_or_path}/talker/hift.pt'
)
with open(f'{model_name_or_path}/talker/audio_detokenizer.yaml', 'r') as f:
 configs = load_hyperpyyaml(f)
audio_frontend = TTSFrontEnd(
 configs["feat_extractor"],
 f'{model_name_or_path}/talker/campplus.onnx',
 f'{model_name_or_path}/talker/speech_tokenizer_v1.onnx',
)
tts_text = "这是一条测试语句。"
spk_input = audio_frontend.frontend_zero_shot(prompt_text="感谢你的认可。", prompt_wav_path="data/spks/prompt.wav")
audio_tokens = model.talker.omni_audio_generation(tts_text, **spk_input)
waveform = audio_detokenizer.token2wav(audio_tokens, save_path='out.wav', **spk_input)
```
For detailed usage for ASR, SpeechQA, and TTS tasks, please refer to `test_audio_tasks.py`
### Image Generation & Edit
Ming-omni natively supports image generation and image editing. To use this function, you only need to add the corresponding parameters in the generate function.
```python
# Image generation mode currently limits the range of input pixels.
gen_input_pixels = 451584
processor.max_pixels = gen_input_pixels
processor.min_pixels = gen_input_pixels
def generate(messages, processor, model, **image_gen_param):
 text = processor.apply_chat_template(messages, add_generation_prompt=True)
 image_inputs, video_inputs, audio_inputs = processor.process_vision_info(messages)
 inputs = processor(
 text=[text],
 images=image_inputs,
 videos=video_inputs,
 audios=audio_inputs,
 return_tensors="pt",
 ).to(model.device)
 for k in inputs.keys():
 if k == "pixel_values" or k == "pixel_values_videos" or k == "audio_feats":
 inputs[k] = inputs[k].to(dtype=torch.bfloat16)
print(image_gen_param)
 image = model.generate(
 **inputs,
 image_gen=True,
 **image_gen_param,
 )
 return image
```
Text-to-image
```python
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "text", "text": "Draw a girl with short hair."},
 ],
 }
]
image = generate(
 messages=messages, processor=processor, model=model,
image_gen_cfg=6.0, image_gen_steps=20, image_gen_width=480, image_gen_height=544
)
image.save("./t2i.jpg")
```
Edit
```python
messages = [
 {
 "role": "HUMAN",
 "content": [
 {"type": "image", "image": "samples/cake.jpg"},
 {"type": "text", "text": "add a candle on top of the cake"},
 ],
 }
]
image = generate(
 messages=messages, processor=processor, model=model,
image_gen_cfg=6.0, image_gen_steps=20, image_gen_width=512, image_gen_height=512
)
image.save("./edit.jpg")
```
## License and Legal Disclaimer
This code repository is licensed under the [MIT License](../LICENSE), and the Legal Disclaimer is located in the [LEGAL.md file](../LEGAL.md) under the project's root directory.
## Citation
If you find our work helpful, feel free to give us a cite.
```bibtex
@misc{Mingomni2025,
 title = {Ming-Omni: A Unified Multimodal Model for Perception and Generation},
author = {Inclusion AI},
 year = {2025},
 eprint = {2506.09344},
 archivePrefix = {arXiv},
 url = {https://arxiv.org/abs/2506.09344}
}
```