Jinshan Office Announces Work with Nvidia to Accelerate WPS AI Landing
On June 30, the official public number of Jinshan Office announced that the company cooperated with the Nvidia team to accelerate the landing of WPS AI.。
In order to chase the wave of big models, many Chinese companies have recently announced cooperation with the global chip giant Nvidia, and Jinshan Office is not far behind.。
On June 30, the official public number of Jinshan Office showed that the company and the Nvidia team reached a cooperation to accelerate the landing of WPS AI.。
As an office software and service provider in China, Jinshan Office's products include WPS Office office software, collaborative office products such as files, and image document recognition and understanding services.。
Among them, Jinshan Office has reached the domestic leading level in the field of image document recognition and understanding.。It mainly uses deep learning technologies such as CV (Computer Vision) and VIE (Visual Information Extraction, Visual Information Extraction), such as identifying invoices and PDF outlines through CV technology and obtaining data relationships in them, with hundreds of millions of requests per day.。
However, when the image recognition and understanding business landed, the company team faced two major problems: absolute time-consuming and high cost.。In the cooperation with Nvidia, these two problems have been effectively solved。
Long links (up to 20 + models of pipeline) mean long time consumption, if based on CPU inference, pipeline time will be up to 15 seconds, through GPU inference and TensorRT acceleration, successfully reduced the time consumption to 2.About 4 seconds。
When deploying models, it is common practice to place the inference and business code in a process space, and each model needs to be loaded once in each process.。Even if the process pool is used to improve the GPU utilization of a single card, the overall GPU utilization of the machine is not high due to CPU constraints.。After the introduction of the Triton inference server for inference deployment, the inference is decoupled from the business code, and the initial implementation of the GPU resource pool scheduling。Finally, at the same business size, the deployment cost savings of 23%。
WPS AI is an artificial intelligence application with large language model capability under Jinshan Office, and is also the first ChatGPT-like application on the domestic collaborative office track.。WPS AI is now connected to WPS text, presentation, form, PDF, intelligent document, intelligent form and other office components under Jinshan Office。
It is reported that Jinshan Office has been doing AI for the past few years, but using traditional small models.。Perhaps can not find a better breakthrough, by the end of last year, Jinshan office has nearly abandoned AI。But the emergence of ChatGPT has brought them back to the AI track.。
"When I saw that AI could write programs on its own, I knew that the world had completely changed.。Jinshan Office CEO Zhang Qingyuan said in an interview in May, "We feel that this will be more violent and disruptive than all the technological changes that Jinshan Office has experienced in its history."。As a software company, we must actively embrace, otherwise we will be eliminated。"
Determined to invest in AI again, Zhang Qingyuan said, considering the cost problem, Jinshan Office decided to do the application of the big model first, and then solve the problem of large model self-control in the future.。
On April 18, Jinshan Office announced that it would launch WPS AI。He also said that WPS AI will be the first to be applied to a new generation of online content collaboration editing tools - light documents to help users deal with work and life scenarios to improve content productivity.。Light documents with WPS AI will unlock a variety of functions, including document generation, multiple rounds of dialogue, processing and reprocessing of existing documents, and quick summarization of document content.。
On May 16, WPS AI was released on the new。WPS AI embeds large model capabilities into four major components: text, tables, presentations, and PDF。WPS AI believes that the new show on the reading comprehension and question-and-answer and human-computer interaction in the general direction of the deep ability。
However, WPS AI is still in the compliance phase and has not yet been officially launched, and WPS AI is only the current product code.。
According to the official introduction, WPS AI will anchor three strategic directions in the future:
The first direction is AIGC, mainly to assist editing, rewriting, polishing, to help users produce better quality articles; the second direction is reading comprehension and question-and-answer, based on the stored corpus to do question-and-answer interactive assisted reading and knowledge retrieval; the third direction is human-computer interaction, based on large language model technology to do the next generation of natural language-based human-computer interaction, using natural human language to command and request software。
Zhang Qingyuan also revealed the next move of Jinshan Office: "On the one hand, it is the overseas version, which is to actively follow up on the world's most advanced technology, and I have to have everything my opponent has, and do all of Copilot, ChatPDF, ChatDOC, and ChatExcel.。On the other hand, the domestic version may be slower, waiting for the domestic partner's big model to go live.。"
At present, WPS AI and minimax, Wenxin big model, intelligent spectrum AI three big language model cooperation, the future will be oriented to office scenarios, explore typical applications.。
·Original
Disclaimer: The views in this article are from the original author and do not represent the views or position of Hawk Insight. The content of the article is for reference, communication and learning only, and does not constitute investment advice. If it involves copyright issues, please contact us for deletion.