Open SourceHigh importance
Baidu releases ERNIE-5.5-Flash-Base weights: MoE 280B (14B active), pre-training-only checkpoint
Baidu published on Thursday (11/05) ERNIE-5.5-Flash-Base on Hugging Face: a sparse MoE with 280B total parameters and 14B active per token, pre-trained from scratch on 33.5T tokens, training context up to 128K and support for 1M tokens in fine-tune/inference. The checkpoint is the end of pre-training, with no mid- or post-training. The company reports, on its own harness against similar bases (DeepSeek-V4-Flash-Base, Hy3-preview-Base, MiMo-V2.5-Base), MMLU 93.3, MMLU-Pro 78.4, HumanEval+ 82.3 and LiveCodeBench 40.1. Weights in FP8 and BF16; custom license ERNIE 5.5 Flash Base Model License Agreement v1.0. The post-trained model “will follow”. The @Baidu_Inc X post had about 214 thousand views and 255 likes in this edition's measurement.
Published in edition no. 12 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
Baidu ERNIEWeights on Hugging FaceOfficial @Baidu_Inc post
https://ernie.baidu.com/release/ernie-5-5-flash-base/