On May 9, Baidu announced the official release of its next-generation foundation model, Ernie Foundation Model 5.1.
According to the introduction, Ernie 5.1 adopts "Multi-Dimensional Elastic Pre-training" technology, achieving leading baseline performance with only about 6% of the pre-training cost of industry models of the same scale, and ranking first domestically on the LMArena Search Leaderboard.
In addition, the latest rankings from the Large Model Arena show that Ernie 5.1 scored 1,223 points, placing first domestically and fourth globally on the LMArena Search Leaderboard, making it the only Chinese model on the list.
The significant improvement in Ernie 5.1's comprehensive capabilities is attributed to key technologies such as "Multi-Dimensional Elastic Pre-training." This technology was first introduced with the release of Ernie 5.0, enabling the generation of multiple model scales from a single training run. As a milestone achievement of this technology, Ernie 5.1 fully inherits the knowledge from Ernie 5.0, compressing the total parameters to approximately 1/3 and the active parameters to approximately 1/2, while achieving leading baseline performance with only about 6% of the pre-training cost of industry models of the same scale.
Currently, Ernie 5.1 has been simultaneously launched on Baidu Qianfan Model Square and the Ernie Bot official website, open for experience by enterprise users and developers.
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (0)