Intel / article

ERNIE 5.1 Officially Released! Topping Multiple Leaderboards — A Model That Writes Better and Understands You More

Source: unknown
Date: Published May 9, 2026

Context

ERNIE 5.1 reuses the training foundation of ERNIE 5.0 to produce a smaller model. Baidu says the earlier training process varied network depth, available computational experts and how many experts were activated. That trained multiple configurations within one larger system, from which the team selected the configuration for 5.1. The release article explains this reuse.

The company reports roughly one-third as many total parameters and half as many active parameters as 5.0. Those figures distinguish the amount of learned capacity stored from the amount used during computation. The supported efficiency story is therefore about extracting a smaller trained configuration and reducing work, rather than training every model size independently.

Terms in this piece · Glossary

inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.

Source link unavailable.unknown