Artificial Intelligence thread

[european_guy]

Please, Log in or Register to view URLs content!

Huawei's play here with this AI cloud that have all the models. everyone needs more comptuation power and AI chips are in hot demand and too expensive and short supply

no problem, use my datacenter here with pre configured models and everything, it will be more efficient.

The article is a bit scarce of technical details. Form this

The multi-cloud collaborative architecture allows industry large models to be trained on the public cloud, fine-tuned on the hybrid cloud based on local enterprise data

One can assume that the basic idea is to use Huawei cloud for pre-training on general, non sensible data. This is the most compute intensive part that can easily take months for big models.

Then the pre-trained model is fine-tuned on private data, possibly even on the corporate premise and on the corporate's own hardware. This part is more then an order of magnitude less compute intensive than pre-training. Usually lasts few days, maybe a couple of weeks if done on customer's limited hardware.

So customers can have their cake and eat it too: pre-training with huge but general data on Huawei cloud and lean fine tuning with private data on private hardware.

For the specific model to use, I'd guess customer can choose an off-the shelf one, maybe even a pre-trained one, or its own original architecture: model architecture does not seem to be an issue as long as code is compatible with Huawei's MindSpore or even Pythorh now that

Please, Log in or Register to view URLs content!

.

 

[BlackWindMnt]

I know that Microsoft with Azure also announced their "Models as a Service" azure product.
Im not completely sure how it works but AI was a big part of their developers event MS ignite they held recently.
Where they pretty much said Microsoft is now an AI company.

 

[tphuang]

https://twitter.com/i/web/status/1730248977499762740

interesting new LLM from China called DeepSeek, 67B parameters

 

[sunnymaxi]

from AliBaba

https://twitter.com/i/web/status/1730127387109781932

We are proud to present our sincere open-source works: Qwen-72B and Qwen-1.8B! Including Base, Chat and Quantized versions!

Qwen-72B has been trained on high-quality data consisting of 3T tokens, boasting a larger parameter scale and more training data to achieve a comprehensive performance upgrade. Additionally, we have expanded the context window length to 32K and enhanced the system prompt capability, allowing users to customize their own AI assistant with just a single prompt.

Qwen-1.8B is our additional gift to the research community, striking a balance between maintaining essential functionalities and maximizing efficiency, generating 2K-length text content with just 3GB of GPU memory.

@tphuang ..

 

[Topazchen]

https://twitter.com/i/web/status/1730248977499762740

interesting new LLM from China called DeepSeek, 67B parameters

So snappy. The next iteration will probably be miles ahead of GPT-4

Please, Log in or Register to view URLs content!

 

[sunnymaxi]

https://twitter.com/i/web/status/1730027392532676785

Chinese enterprises generally have a high acceptance of generative

Please, Log in or Register to view URLs content!

. Currently, 67% of Chinese enterprises have started exploring opportunities for the application of generative AI within their organizations or have begun substantial investments in this area..

Chinese firms implementing Ai into manufacturing process.

 

[Overbom]

Huge compute advantage for Meta and Microsoft

 

[Arij Javaid]

When will China overtake Nvidia in A.I chips??

 

[measuredingabens]

When will China overtake Nvidia in A.I chips??

Unknown. There are methods that could potentially allow Huawei (which is really the only viable Chinese competitor to Nvidia) to leapfrog Nvidia's best offerings, though for now they are playing catchup and probably staying with conventional AI chip setups. Huawei does have a massive advantage in networking that allows it to compete on a per-setup basis, though chip wise it still has a ways to go.

 

[gelgoog]

China will have a lag in terms of chip fabrication technology at least in the near term. So whoever wants to design an AI chip in China will have to make an hardware architecture more optimized for the specific use case of AI instead of just lightly modifying a GPU like NVIDIA does. Unfortunately this will mean it will be less suitable for general purpose uses.