Artificial Intelligence thread

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
I wouldn't necessarily call it second rate. It's actually better than Hopper in terms of some performance metrics and memory bandwidth. Which makes sense, since it's sampling only in the next few months, significantly later than Hopper. By the time MI300 is shipping, Hopper will have been out for a year.

View attachment 114463
One of the main issue with MI300 is a lack of a competitive software stack.

But Dylan's point is that Chinese firms are actually good at providing their own software solutions, rather than relying on Nvidia's, which might make MI300 an attractive product for them.

Another major issue is "inferior" networking.



So Idk, I guess you guys would have to tell me.
That's not the issue with it. The issue with it is CUDA. You need to come in either significantly better or at a price point lower than Nvidia GPUs to displace it. There are a lot of customers in China who probably support CUDA and not optimized for any other non-Chinese GPUs. China itself also has and will have more GPUs that are competitive with H100. You can look up the stuff I posted on Ascend clusters on this thread.
 

Strangelove

Colonel
Registered Member
Please, Log in or Register to view URLs content!

Chinese companies at cutting-edge of artificial intelligence, says report​


By FAN FEIFEI | CHINA DAILY | Updated: 2023-06-13 07:32


6487bb07a31033ad342ae8ea.jpeg
Visitors look at core components for AI foundation models during an expo in Beijing. CHEN XIAOGEN/FOR CHINA DAILY

Chinese enterprises are at a crucial juncture of breakthroughs in artificial intelligence, as generative AI and other rapidly evolving technologies usher in a future for business where the physical and digital worlds are inextricably linked, global consultancy Accenture said in a report.

According to the report, 91 percent of the business executives surveyed in the country agreed that AI foundation models will play an important role in their organization's strategies over the next three to five years.

About 80 percent of business leaders believed advancements in generative AI are opening a new era of intelligence for enterprises, while 77 percent said AI models will enable connections across various types of data, revolutionizing where and how AI is used.

The research surveyed 225 Chinese companies in 19 industries, including retail, software, industrial goods and equipment, healthcare, energy and financial services from December 2022 to January 2023.

Generative AI refers to algorithms that are trained with huge amounts of data and capable of generating content such as images, text, audio and video. It is the key technology underpinning ChatGPT, an AI chatbot developed by US-based AI research company OpenAI.

The meteoric rise of ChatGPT has captivated the world's attention on the power of generative AI to augment human capability, the report said.

Accenture estimates that as much as 40 percent of all working hours will be supported or augmented by language-based AI.

"The next decade will be defined by three mega technology trends — cloud, metaverse and AI — which collectively will collapse the distance of our digital and physical worlds," said Yu Yi, technology lead for Accenture China.

"As generative AI will have far-reaching impact, business leaders must act immediately and scale up investments in data, talent and customized AI models to meet the unique needs of enterprises," Yu said.

Most Chinese executives agreed that generative AI will spark significant creativity and innovation and usher in a new era of enterprise intelligence, Yu added.

They also said their companies expect quick large-scale analysis abilities and accelerated innovation to be two major benefits from the use of AI models.

Generative AI is expected to usher in speedy growth in China and inject fresh momentum into a new round of industrial revolution, as major Chinese tech heavyweights step up efforts to develop their own AI-powered large models, experts said.

Alibaba Group unveiled its AI-powered large language model Tongyi Qianwen in April, a potential rival to ChatGPT. In March, Baidu Inc rolled out its large language model and Chinese-language ChatGPT alternative, Ernie Bot, which can be used for a variety of functions including searches, autonomous driving and smart devices.

"We believe the emergence of generative AI and large language models presents a transformative potential of AI in various industries to help people and organizations achieve more and have a positive impact on society," said Robin Li, co-founder and CEO of Chinese tech giant Baidu.

Chinese tech companies have the advantage of offering users AI-powered conversational results based on Chinese language and culture, said Lu Yanxia, research director at IDC China.

"Generative AI has brought new business opportunities for domestic AI servers, cloud computing and chip companies," Lu said.

AI and ChatGPT-related technologies are likely to free humans from tedious tasks and enable them to focus on creative thinking, resulting in significant application potential in fields such as culture, retail, finance, healthcare and education, she added.

Data from global market research company International Data Corp show that China's spending on AI will likely hit $14.75 billion this year, accounting for about one-tenth of the global market. The scale of China's AI market is expected to reach $26.44 billion in 2026, with a compound annual growth rate of more than 20 percent from 2021 to 2026.
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Huawei is having a major Ascend AI conference. In there, CAS unveiled 紫东太初 2.0 LLM
Please, Log in or Register to view URLs content!
When 1.0 was announced in Sep 2021, it was world's 1st 100 billion parameter LLM. So, this should be even greater in scope. Trained using Ascend/MindSpore platform
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Huawei is having a major Ascend AI conference. In there, CAS unveiled 紫东太初 2.0 LLM
Please, Log in or Register to view URLs content!
When 1.0 was announced in Sep 2021, it was world's 1st 100 billion parameter LLM. So, this should be even greater in scope. Trained using Ascend/MindSpore platform
This was released as part of 人工智能框架生态峰会 2023 (AI ecosystem framework conference)

Aside from 紫东太初 2.0, it had unveiling of Shanghai Mindspore AI framework & big model innovation center
Please, Log in or Register to view URLs content!

Mindspore (HW) joined hands with 18 top AI institutions to 共建人工智能框架生态倡议 (collectively build AI framework ecosystem agreement)
Already 22 customers have joined the platform this includes
上海人工智能实验室、上海交通大学、中国商飞上海飞机设计研究院、中国科学院上海药物研究所、云从科技
so COMAC, Shanghai jiaoda, Shanghai AI lab, CAS Shanghai medical research, Yuncang technology

截至2023年5月底,昇思开源社区下载量已突破474万,社区贡献者1.3万,服务企业数量5500家,与240家科研院所展开合作,开源模型达400多个,发表顶会论文超过900篇
so 4.74 million download of MindSpore software, 13000 community contributors, 5500 companies & 240 research institutes with 400+ open source models & 900+ research paper

Please, Log in or Register to view URLs content!
上海人工智能研究院、上海市闵行区政府、临港集团、昇思开源社区联合启动上海昇思AI框架&大模型创新中心(简称创新中心),创新中心将携手产业伙伴基于昇思MindSpore AI框架,支持全国范围的AI技术企业、高校与科研院所孵化大模型、科学智能技术研究,进而赋能企业打造极简易用、极致性能的AI解决方案和应用,服务互联网、电信运营商、金融、教育、医疗等人工智能关键产业方向,打造一系列富有竞争力的“AI+行业”的示范性应用场景,推动产业集聚。目前首批22家单位正式入驻创新中心,将联合昇思社区进行大模型创新与AI产学研转化,成为AI技术与应用创新的“发源地”,共同推动人工智能框架生态高质量发展迈出更大步伐。
Looks like Shanghai AI lab, Lingang group, Shanghai govt, Mindspore open source community joined hand to form this.
Since I have not seen similar announcement with other big tech, I can only assume Chinese gov't & major SOEs have coalesced around Huawi AI technology for various industries. Obviously, they use it to develop their own proprietary models, but HW tech is what they use.

Please, Log in or Register to view URLs content!
根据华为官方消息,昇思已使能孵化超20个大模型,到年底希望使能孵化40~50个大模型。昇思已孵化的大模型包括悟道2.0、紫东·太初、秦岭·翔语、CodeGeeX等基础大模型,以及金陵·女娲、鹏程·神农、空天·灵眸等行业大模型。
Already over 20 large models developed using Mindspore, hoping to reach 40 to 50 large modesl by end of this year
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Kind of interesting day

I saw tencent offering large model resources as service calling it Maas

Alibaba is doing their own llm with ysginhua university

JD will announce their ai for industries framework on 07/17

Changhong tv also unveiled their llm for listening, hearing and other sensories.

A lot of movement. Every tech company seems to be doing llm of some type. Btw, I am not sure it's a good idea for everyone to be rushing in.
 
Top