Is really impressive how China is putting these AI models and techniques to practical industrial use.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Alibaba open sourced another Qwen LLM with 32B parameters, now 7 open sourced LLMs
AI in engineering material testing equipment.
Total investment is 600 million yuan! Advantech artificial intelligence testing equipment production base starts construction in Pukou, Nanjing.
The groundbreaking ceremony of Advantech's artificial intelligence testing equipment production base project was held in Nanjing Pukou Economic Development Zone.
It is reported that the project has a total investment of 600 million yuan and covers an area of 37 acres.
Nanjing Advantech Intelligent Technology Co., Ltd. is a company that has been deeply involved in the field of intelligent testing equipment for many years. The artificial intelligence testing equipment production base project started this time has a total construction area of approximately 38,000 square meters. Once completed, the project will help Nanjing Advantech become an industrial digital intelligence benchmark enterprise with leading technology and practice, and accelerate the digital transformation of the building materials testing industry. process. After the project is completed and put into operation, it is expected to achieve revenue of 300 million yuan and tax revenue of 13.3 million yuan in the first year.
An update on just how much computation power is needed to train something complicated like ADAS.
Huawei has 3.3EFLOPS of cloud computation for ADAS and is able to consume 12 million km per day and complete training every 5 days
Here's the full report where that chart was taken:
Alibaba-backed Moonshot AI claims breakthrough in expanded Chinese-character prompt for Kimi chatbot
Moonshot AI’s updated Kimi chatbot can handle up to 2 million Chinese characters in a single prompt, up from the previous 200,000 characters
Company founder Yang Zhilin has said that the ability to process long prompts is critical to the future development of AI models
Llama 3 release is a big deal for the open source community. It's 7B model could run in lots of people's computers and its 70B model could be run by the AI community who care more about AI applications.
And finally, the 400B (still training) model could (i assume) still be ran by some kinda rich people when Apple releases the next Mac Studio lineup (rumoured maximum of half terabyte kinda VRAM).
Meta and Zuckerberg are cooking something big here with this release. Llama 3 blows everything else (open and close source) out of the water when you take into account parameter size. Only disappointing thing is is limited context length, but since they promised that they will release with longer context length in the following months, let's wait
And finally, the 400B (still training) model could (i assume) still be ran by some kinda rich people when Apple releases the next Mac Studio lineup (rumoured maximum of half terabyte kinda VRAM).
Meta and Zuckerberg are cooking something big here with this release. Llama 3 blows everything else (open and close source) out of the water when you take into account parameter size. Only disappointing thing is is limited context length, but since they promised that they will release with longer context length in the following months, let's wait
Alibaba cloud is first in China to offer Llama 3 training and inference
Sensetime releases SenseNova 5.0, aimed at GPT4
Claims a bunch of rankings doing better than GPT4
has 12000 PFLOPS computation now and will have 18000 PFLOPS by end of year with 2000 PLFOPS using domestic chips
Sensetime releases SenseNova 5.0, aimed at GPT4
Claims a bunch of rankings doing better than GPT4
has 12000 PFLOPS computation now and will have 18000 PFLOPS by end of year with 2000 PLFOPS using domestic chips