Baidu has released an end-of-pre-training checkpoint under a custom licence.
A base checkpoint supports further training rather than serving as a finished assistant.
What happened
Baidu describes ERNIE 5.5 Flash Base as a 280B-Parameters: The learned numbers inside a model. More parameters usually means a more capable but larger model. Glossary sparse mixture-of-experts model with Mixture of experts (active parameters): A design where only part of the model (the active parameters) works on each piece of text. Glossary parameters, pre-trained on 33.5T Tokens: The small chunks of text a model reads and writes, often a word or part of a word. Glossary. The released checkpoint is the end of pre-training, without mid- or post-training. The weights use a custom licence; consult the model tracker for the conditions.
Why it matters: A base checkpoint supports further training rather than serving as a finished assistant.
Sources: Primary source, “ERNIE 5.5 Flash Base · ERNIE”; Primary source, “baidu/ERNIE-5.5-Flash-Base · Hugging Face”; Primary source, “LICENSE · baidu/ERNIE-5.5-Flash-Base at main”