Baidu Intelligent Cloud Qianfan large model platform is the first to adapt to Falcon-180B!

fe7e8f7de4e5d0b1ae67d30b20d81d28.gif

Recently, Falcon-180B, as the largest open source model with the largest parameters, was officially released and supported for free commercial use, attracting widespread attention in the industry. Baidu Intelligent Cloud Qianfan large model platform has been adapted to Falcon-180B for the first time, and provides a full-process tool chain and development environment for model fine-tuning, deployment, and application development, making it convenient for enterprises and developers to take the lead in experiencing it.

It is understood that Falcon-180B has shown super performance in various tasks such as reasoning, coding, proficiency and knowledge testing, and it topped the Hugging Face rankings as soon as it was released. In order to facilitate developers to apply the latest large model capabilities, Baidu Smart Cloud Qianfan platform has adapted Falcon-180B for the first time. Developers and enterprises can now apply for experience in the platform's "Model Warehouse". Developers can also add their own Models are imported into the platform to use the full functionality of the platform.

c6c05d109f3a0de478fc43e1147f34b6.png

31bcc8d863e00a904db92ee34ad0f7de.png

Developers can import any model in Hugging Face Transformer format by themselves

It is understood that in order to meet the diverse model needs of developers and enterprises, Baidu Intelligent Cloud Qianfan platform has previously been connected to Falcon-7b and Falcon-40b-instruct. Up to now, the platform has hosted 42 domestic and foreign mainstream large models, including Falcon, Llama 2 full series, RWKV, and Stability Diffusion, ranking first in the number of domestic models.

Baidu Smart Cloud stated that the Baidu Smart Cloud Qianfan platform is not only a simple access, but also performs secondary performance enhancements, security enhancements, Chinese enhancements and context enhancements for third-party models. For example, in terms of performance enhancement, the overall throughput of training Llama2 can be increased by 25%, and the inference performance can even be increased by 2 times.

>>In terms of security, which enterprise customers are most concerned about , Baidu Intelligent Cloud Qianfan platform provides multiple guarantees of model security, content security and data security. This includes designing dedicated isolation domains to protect the security of model assets; pre-configuring security modules to effectively control input and output content risks; and providing full-link encryption from models, data to logs to ensure data security. A three-pronged approach allows both model manufacturers and customers to enjoy the security of the platform.

For most enterprises, there is still a high threshold for model development and deployment. In order to help enterprises make good use of large models, Baidu Intelligent Cloud Qianfan's tool chain runs through the complete life cycle of large models. Based on the one-stop visualization tool, the time for customers to verify the effect of the model is greatly shortened. They can run through a model and see the effect in one day.

>>In terms of data , in order to solve the problem of high quality and high cost of enterprise model training data, Baidu Intelligent Cloud Qianfan platform also presets 41 high-quality and industry-specific data sets, which developers can complete with the click of a few buttons. Fine-tune and improve model effects, further lowering the threshold for enterprises to use, train, and infer large models.

>>In terms of computing power , when Baidu Intelligent Cloud Qianfan Platform trains large models on the Wanka cluster, the effective training time of the model reaches more than 95%. The acceleration ratio in Wanka-scale clusters can reach 95%, fully releasing the overall computing power of the cluster and significantly reducing customers' computing power costs and time costs. It is worth mentioning that the platform is also compatible with domestic and foreign mainstream chips and operating systems, and customers’ algorithms can be run with minimal modification costs.

8ea9fc1f6a4fc3e57c0c209d59996bb8.png

Baidu Intelligent Cloud Qianfan Large Model Platform Product Panorama

At the just-concluded 2023 Baidu Cloud Intelligence Conference, Baidu Intelligent Cloud Qianfan large model platform achieved a comprehensive 2.0 upgrade. As of now, the number of monthly active companies on the platform has reached nearly 10,000, covering more than 400 scenarios in finance, manufacturing, energy, government affairs, transportation and other industries.

Shen Dou, executive vice president of Baidu Group and president of Baidu Intelligent Cloud Business Group, said, "The real implementation of any new technology requires the support of corresponding methodologies and tools." "We spent half a year, constantly accumulating experience, accumulating tools, and Baidu Intelligent Cloud Qianfan platform has made a lot of upgrades to make the large model platform more usable and considerate."

e2d93a27d7fc0613e39c414be4621f88.png

1dec0aa51ca9615d1f1f27f5aa21c51c.png

92ec4514dca7e356025a2fbffe327876.jpeg

196ed971ef44c516b6e4ac2de447bc52.jpeg

3d0dbf8816453a8cdd206581c47dae6e.jpeg

Click "Read the original text" to experience it immediately!

Guess you like

Origin blog.csdn.net/weixin_48493350/article/details/132769820