news

image source: visual china

author: an ran

editor: ye jinyan

produced by: deepnet·tencent news xiaoman studio

on september 2, 2024, infinigence ai announced the completion of nearly 500 million yuan in series a financing. so far, in just 1 year and 4 months since its establishment, infinigence ai has completed a total of nearly 1 billion yuan in financing.

the joint lead investors in this round of financing for wuwen xinqiong are the social security fund zhongguancun independent innovation special fund, qiming venture partners and hongtai fund. the follow-up investors include strategic investors such as lenovo capital, xiaomi, and softbank hi-tech, state-owned funds such as china development bank science and technology innovation, shanghai artificial intelligence industry investment fund, and xuhui science and technology investment, as well as financial institutions such as shunwei capital, dachen capital, detong capital, shangshi capital, senruo yukun, shenwan hongyuan, and zhengjing capital.

in this regard, xia lixue, co-founder and ceo of wuwen core qiong, said: "the new '80/20 rule' brought by the ai ​​2.0 wave, the transformer architecture has unified the new technical paradigm, which means that only 20% of the key technical problems need to be solved to support 80% of the generalization of vertical scenarios, providing a rare opportunity for the standardization and scale-up of software and hardware joint optimization technology; the contradiction between supply and demand and the uneven distribution of resources facing the chinese computing power ecosystem have created an era opportunity for us to drive upstream and downstream cooperation to achieve efficient integration of diversified heterogeneous computing power."

the founding team of wuwen xinqiong is from the department of electronic engineering of tsinghua university. the team initially determined that the actual available computing power of large models not only depends on the theoretical computing power of the chip, but also can amplify the computing power utilization efficiency through the optimization coefficient and amplify the overall computing power scale through the cluster scale. therefore, wuwen xinqiong proposed the formula of "chip computing power × optimization coefficient (software and hardware collaboration) × cluster scale (multi-heterogeneous) = ai model computing power".

following this formula, wuwen xinqiong will continue to improve the utilization rate of chip computing power in large-model tasks through joint software and hardware optimization technology, and improve the utilization rate of cluster computing power through multi-heterogeneous computing power adaptation technology to expand the industry's overall computing power supply.

it should be pointed out that different hardware platforms need to adapt to different software stacks and tool chains, and there has long been an "ecological silo" phenomenon between heterogeneous chips that is difficult to use.

in terms of joint optimization of software and hardware, wuwen xinqiong has greatly improved the utilization rate of mainstream hardware and heterogeneous hardware through its self-developed inference acceleration technology flashdecoding++, surpassing the prior sota and completing the adaptation of multiple mainstream open source large models on more than 10 computing cards including amd, huawei ascend, biren, cambrian, suiyuan, haiguang, tianshu zhixin, muxi, moore thread, nvidia, etc.

in terms of software and hardware collaboration and multiple heterogeneous technologies, wuwen core qiong has built the infini-ai heterogeneous cloud platform based on the multiple chip computing power base. the infini-ai heterogeneous cloud platform also includes a one-stop ai platform (aistudio) and a large model service platform (genstudio).

among them, aistudio's one-stop ai platform provides machine learning developers with cost-effective development and debugging, distributed training, and high-performance reasoning tools, covering the entire life cycle from data hosting, code development, model training, and model deployment.

the genstudio large model service platform provides high-performance, easy-to-use, safe and reliable multi-scenario large model services for large model application developers, covering the entire process from large model development to service-oriented deployment, effectively reducing development costs and thresholds.

wuwen xinqiong officials stated that by activating multi-heterogeneous computing power and joint optimization of software and hardware, wuwen xinqiong aims to reduce the implementation cost of large models by 10,000 times, just like "water, electricity and coal", becoming a new type of productivity that is within the reach of the industry and widely benefits the industry, accelerating the process of universal agi.