Large multi-modal models for strong performance and efficient deployment
MiniCPM-V and OmniLMM are a family of open-source large multimodal models (LMMs) adept at vision & language modeling. The models process images and text inputs and deliver high-quality text outputs. We release two featured versions that are targeted at strong performance and efficient deployment:
-
MiniCPM-V 2.8B: State-of-the-art end-side large multimodal models. Our latest MiniCPM-V 2.0 can accept 1.8 million pixels (e.g., 1344x1344) images at any aspect ratio, and is adept at OCR capability. It achieves comparable performance with Gemini Pro in understanding scene-text and matches GPT-4V in preventing hallucinations.
-
OmniLMM 12B: The most capable version with leading performance among comparable-sized models on multiple benchmarks. The model also achieves state-of-the-art performance in trustworthy behaviors, with even less hallucination than GPT-4V.
MiniCPM-V 2.8B is an efficient version with promising performance for deployment. The model is built based on SigLip-400M and MiniCPM-2.4B, connected by a perceiver resampler. Our latest version, MiniCPM-V 2.0 has several notable features.
-
🔥 State-of-the-art Performance.
MiniCPM-V 2.0 achieves state-of-the-art performance on multiple benchmarks (including OCRBench, TextVQA, MME, MMB, MathVista, etc) among models under 7B parameters. It even outperforms strong Qwen-VL-Chat 9.6B, CogVLM-Chat 17.4B, and Yi-VL 34B on OpenCompass, a comprehensive evaluation over 11 popular benchmarks. Notably, MiniCPM-V 2.0 shows strong OCR capability, achieving comparable performance to Gemini Pro in scene-text understanding, and state-of-the-art performance on OCRBench among open-source models.
-
🏆 Trustworthy Behavior.
LMMs are known for suffering from hallucination, often generating text not factually grounded in images. MiniCPM-V 2.0 is the first end-side LMM aligned via multimodal RLHF for trustworthy behavior (using the recent RLHF-V [CVPR'24] series technique). This allows the model to match GPT-4V in preventing hallucinations on Object HalBench.
-
🌟 High-Resolution Images at Any Aspect Raito.
MiniCPM-V 2.0 can accept 1.8 million pixels (e.g., 1344x1344) images at any aspect ratio. This enables better perception of fine-grained visual information such as small objects and optical characters, which is achieved via a recent technique from LLaVA-UHD.
-
⚡️ High Efficiency.
MiniCPM-V 2.0 can be efficiently deployed on most GPU cards and personal computers, and even on end devices such as mobile phones. For visual encoding, we compress the image representations into much fewer tokens via a perceiver resampler. This allows MiniCPM-V 2.0 to operate with favorable memory cost and speed during inference even when dealing with high-resolution images.
-
🙌 Bilingual Support.
MiniCPM-V 2.0 supports strong bilingual multimodal capabilities in both English and Chinese. This is enabled by generalizing multimodal capabilities across languages, a technique from VisCPM [ICLR'24].
Click to view results on TextVQA, DocVQA, OCRBench, OpenCompass, MME, MMBench, MMMU, MathVista, LLaVA Bench, Object HalBench.
Model | Size | TextVQA val | DocVQA test | OCRBench | OpenCompass | MME | MMB dev(en) | MMB dev(zh) | MMMU val | MathVista | LLaVA Bench | Object HalBench |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Proprietary models | ||||||||||||
Gemini Pro Vision | - | 74.6 | 88.1 | 680 | 63.8 | 2148.9 | 75.2 | 74.0 | 48.9 | 45.8 | 79.9 | - |
GPT-4V | - | 78.0 | 88.4 | 645 | 63.2 | 1771.5 | 75.1 | 75.0 | 53.8 | 47.8 | 93.1 | 86.4 / 92.7 |
Open-source models 6B~34B | ||||||||||||
Yi-VL-6B | 6.7B | 45.5* | 17.1* | 290 | 49.3 | 1915.1 | 68.6 | 68.3 | 40.3 | 28.8 | 51.9 | - |
Qwen-VL-Chat | 9.6B | 61.5 | 62.6 | 488 | 52.1 | 1860.0 | 60.6 | 56.7 | 37.0 | 33.8 | 67.7 | 56.2 / 80.0 |
Yi-VL-34B | 34B | 43.4* | 16.9* | 290 | 52.6 | 2050.2 | 71.1 | 71.4 | 45.1 | 30.7 | 62.3 | - |
DeepSeek-VL-7B | 7.3B | 64.7* | 47.0* | 435 | 55.6 | 1765.4 | 74.1 | 72.8 | 38.3 | 36.8 | 77.8 | - |
TextMonkey | 9.7B | 64.3 | 66.7 | 558 | - | - | - | - | - | - | - | - |
CogVLM-Chat | 17.4B | 70.4 | 33.3* | 590 | 52.5 | 1736.6 | 63.7 | 53.8 | 37.3 | 34.7 | 73.9 | 73.6 / 87.4 |
Open-source models 1B~3B | ||||||||||||
DeepSeek-VL-1.3B | 1.7B | 58.4* | 37.9* | 413 | 46.0 | 1531.6 | 64.0 | 61.2 | 33.8 | 29.4 | 51.1 | - |
MobileVLM V2 | 3.1B | 57.5 | 19.4* | - | - | 1440.5(P) | 63.2 | - | - | - | - | - |
Mini-Gemini | 2.2B | 56.2 | 34.2* | - | - | 1653.0 | 59.8 | - | 31.7 | - | - | - |
MiniCPM-V | 2.8B | 60.6 | 38.2 | 366 | 47.6 | 1650.2 | 67.9 | 65.3 | 38.3 | 28.9 | 51.3 | 78.4 / 88.5 |
MiniCPM-V 2.0 | 2.8B | 74.1 | 71.9 | 605 | 55.0 | 1808.6 | 69.6 | 68.1 | 38.2 | 38.7 | 69.2 | 85.5 / 92.2 |
We deploy MiniCPM-V 2.0 on end devices. The demo video is the raw screen recording on a Xiaomi 14 Pro without edition.
Please see the info about MiniCPM-V 1.0 here.
OmniLMM-12B is the most capable version. The model is built based on EVA02-5B and Zephyr-7B-β, connected with a perceiver resampler layer, and trained on multimodal data in a curriculum fashion. The model has three notable features:
-
🔥 Strong Performance.
OmniLMM-12B achieves leading performance among models with comparable sizes, surpassing established LMMs on multiple benchmarks (including MME, MMBench, SEED-Bench, etc). The model also endows rich multi-modal world knowledge.
-
🏆 Trustworthy Behavior.
LMMs are known for suffering from hallucination, often generating text that is not factually grounded in images (e.g., faithfully describing non-existing objects in images). OmniLMM-12B is the first state-of-the-art open-source LMM aligned via multimodal RLHF for trustworthy behavior (using the recent RLHF-V technique). It ranks #1 among open-source models on MMHal-Bench, and outperforms GPT-4V on Object HalBench.
-
🕹 Real-time Multimodal Interaction.
We combine the OmniLMM-12B and GPT-3.5 (text-only) into a real-time multimodal interactive assistant. The assistant accepts video streams from the camera and speech streams from the microphone and emits speech output. While still primary, we find the model can replicate some of the fun cases shown in the Gemini Demo video, without any video edition.
Click to view results on MME, MMBench, MMMU, MMBench, MMHal-Bench, Object HalBench, SeedBench, LLaVA Bench, MathVista.
Model | Size | MME | MMB dev (en) | MMMU val | MMHal-Bench | Object HalBench | SeedBench-I | MathVista | LLaVA Bench |
---|---|---|---|---|---|---|---|---|---|
GPT-4V† | - | 1771.5 | 75.1 | 56.8 | 3.53 / 70.8 | 86.4 / 92.7 | 71.6 | 47.8 | 93.1 |
Qwen-VL-Plus† | - | 2183.4 | 66.2 | 45.2 | - | - | 65.7 | 36.0 | 73.7 |
Yi-VL 6B | 6.7B | 1915.1 | 68.6 | 40.3 | - | - | 67.5 | 28.8 | 51.9 |
Qwen-VL-Chat | 9.6B | 1860.0 | 60.6 | 35.9 | 2.93 / 59.4 | 56.2 / 80.0 | 64.8 | 33.8 | 67.7 |
CogVLM-Chat | 17.4B | 1736.6 | 63.7 | 32.1 | 2.68 / 52.1 | 73.6 / 87.4 | 68.8 | 34.7 | 73.9 |
LLaVA 1.5 | 13.6B | 1808.4 | 68.2 | 36.4 | 2.71 / 51.0 | 53.7 / 77.4 | 68.1 | 26.4 | 64.6 |
OmniLMM-12B | 11.6B | 1935.8 | 71.6 | 40.7 | 3.45 / 68.8 | 90.3 / 95.5 | 71.1 | 34.9 | 72.0 |
We combine the OmniLMM-12B and GPT-3.5 (text-only) into a real-time multimodal interactive assistant. Video frames are described in text using OmniLMM-12B, and ChatGPT 3.5 (text-only) is employed to generate response according to the descriptions and user prompts. The demo video is a raw recording without edition.
OmniLMM_video_demo_en.mp4
Click here to try out the Demo of MiniCPM-V 2.0 and OmniLMM-12B.
- Clone this repository and navigate to the source folder
git clone https://github.com/OpenBMB/MiniCPM-V.git
cd MiniCPM-V
- Create conda environment
conda create -n MiniCPM-V python=3.10 -y
conda activate MiniCPM-V
- Install dependencies
pip install -r requirements.txt
Model | Description | Download Link |
---|---|---|
MiniCPM-V 2.0 | The latest version for state-of-the-art end-side capabilities with high efficiency. | 🤗 |
MiniCPM-V | The first version of MiniCPM-V. | 🤗 |
OmniLMM-12B | The most capable version with leading performance. | 🤗 |
Please refer to the following codes to run MiniCPM-V
and OmniLMM
.
import torch
from chat import OmniLMMChat, img2base64
torch.manual_seed(0)
chat_model = OmniLMMChat('openbmb/MiniCPM-V-2.0') # or 'openbmb/OmniLMM-12B'
im_64 = img2base64('./assets/hk_OCR.jpg')
# First round chat
msgs = [{"role": "user", "content": "Where should I go to buy a camera?"}]
inputs = {"image": im_64, "question": json.dumps(msgs)}
answer = chat_model.chat(inputs)
print(answer)
# Second round chat
# pass history context of multi-turn conversation
msgs.append({"role": "assistant", "content": answer})
msgs.append({"role": "user", "content": "Where is this store in the image?"})
inputs = {"image": im_64, "question": json.dumps(msgs)}
answer = chat_model.chat(inputs)
print(answer)
We can obtain the following results:
"You should go to the Canon store for a camera."
"The Canon store is located on the right side of the image."
Click to view an example, to run MiniCPM-V 2.0 on 💻 Mac with MPS (Apple silicon or AMD GPUs).
# test.py
import torch
from PIL import Image
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained('openbmb/MiniCPM-V-2.0', trust_remote_code=True, torch_dtype=torch.bfloat16)
model = model.to(device='mps', dtype=torch.float16)
tokenizer = AutoTokenizer.from_pretrained('openbmb/MiniCPM-V-2.0', trust_remote_code=True)
model.eval()
image = Image.open('./assets/hk_OCR.jpg').convert('RGB')
question = 'Where is this photo taken?'
msgs = [{'role': 'user', 'content': question}]
answer, context, _ = model.chat(
image=image,
msgs=msgs,
context=None,
tokenizer=tokenizer,
sampling=True
)
print(answer)
Run with command:
PYTORCH_ENABLE_MPS_FALLBACK=1 python test.py
Currently MiniCPM-V 2.0 can be deployed on mobile phones with Android and Harmony operating systems. 🚀 Try it out here.
- Fine-tuning support
- Local Web-UI deployment
- Code release for real-time interactive assistant
The code in this repo is released according to Apache-2.0
The usage of MiniCPM-V's and OmniLMM's parameters is subject to "General Model License Agreement - Source Notes - Publicity Restrictions - Commercial License"
The parameters are fully open to academic research
Please contact cpm@modelbest.cn to obtain written authorization for commercial uses. Free commercial use is also allowed after registration.
As LMMs, OmniLMMs generate contents by learning a large amount of multimodal corpora, but they cannot comprehend, express personal opinions or make value judgement. Anything generated by OmniLMMs does not represent the views and positions of the model developers
We will not be liable for any problems arising from the use of OmniLMM open source models, including but not limited to data security issues, risk of public opinion, or any risks and problems arising from the misdirection, misuse, dissemination or misuse of the model.
This project is developed by the following institutions: