The open source MiniCPM 2.0 series of models from Mianbi Intelligent has significantly enhanced its OCR and other capabilities

Wall-facing intelligenceup to dateThe new generation of flagship edge-side models launched by the company - the MiniCPM2.0 series models bring a series of amazing performance and features:

1. MiniCPM-V2.0 is on the end sideStrongestofMultimodal Model, with powerful OCR capabilities, even comparable to Gemini Pro in some aspects. It uses self-developed high-definition image decoding technology to accurately recognize various complex image contents, including street scenes and long images.

2. MiniCPM-1.2B is a base model that is more suitable for edge scenarios. Its performance exceeds many mainstream models, including Llama2-13B. Its inference speed is nearly 25 times the human speaking speed, and its cost is also greatly reduced.

3. MiniCPM-2B-128K is currently the smallest long text model, which can process 128K (200,000 words) of text content and performs excellently on the multi-dimensional long text evaluation set.

4. MiniCPM-MoE-8x2B is a MoE architecture model with further enhanced performance, with an average performance improvement of 4.5 percentage points and an inference cost of only 69.7% of Gemini-7B.

The open source MiniCPM 2.0 series of models from Mianbi Intelligent has significantly enhanced its OCR and other capabilities

These new-generation MiniCPM models have demonstrated powerful performance and functionality in different fields and scenarios, promoting the further development of large models in terminal applications. At the same time, Mianbi Intelligence has just completed a new round of financing of hundreds of millions of yuan, and plans to continue the journey of efficient large models for AGI, and welcomes outstanding talents to join their team.

MiniCPM-V2.0:

https://github.com/OpenBMB/MiniCPM-V

MiniCPM SeriesOpen Sourceaddress:

https://github.com/OpenBMB/MiniCPM

MiniCPMTechnical blog address:

https://openbmb.vercel.app/?category=Chinese+Blog

statement:The content is collected from various media platforms such as public websites. If the included content infringes on your rights, please contact us by email and we will deal with it as soon as possible.
Information

Sora replacement? 2-minute long AI video model StreamingT2V free open source trial address announced

2024-4-16 9:43:32

Information

KPMG: 36% executives in the financial sector use generative AI every day

2024-4-16 9:47:29

Search