Baidu Releases ERNIE 5.5 Flash Base Language Model
Baidu has released ERNIE 5.5 Flash Base, a base checkpoint designed for teams looking to further train the model themselves. The release was shared via a Reddit post pointing to the official documentation and model collection. This release provides developers and researchers with a new base model checkpoint from Baidu's ERNIE ecosystem for continued pre-training and customization. It expands the availability of advanced large language model architectures to the open community. The model is a sparse Mixture-of-Experts (MoE) architecture featuring 280 billion total parameters with 14 billion activated per token, trained on 33.5 trillion tokens. It utilizes specialized mechanisms like MLA-DSA layers to maintain strong performance across long-context benchmarks.
## BACKGROUND
Baidu's ERNIE series represents a major family of large language models developed in China, competing globally in natural language processing and multimodal tasks. Base checkpoints are raw model weights provided before instruction tuning, allowing developers to adapt them for specific downstream applications.