Baidu open-sources ERNIE-5.5-Flash-Base foundation model
What happened
On November 5, a Hacker News front-page report said Baidu released the weights of ERNIE-5.5-Flash-Base, a sparse MoE foundation model with 280B total parameters and 14B active per token, pretrained from scratch on 33.5T tokens. Training context extends to 128K, with 1M-token context supported at fine-tuning and inference. The released checkpoint has no mid-training or post-training. On November 6, The Decoder confirmed the open-source release, the parameters and the pretraining scale, and said the design supports scaling to 1 million tokens during fine-tuning and inference. Baidu plans to provide post-trained versions later.
Written by AI from the coverage · updated 5 hours ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderBaidu releases ERNIE 5.5 Flash Base with 280 billion parameters as an open-source foundation model
百度发布开源基础模型 ERNIE-5.5-Flash-Base，未经微调或后训练，计划后续提供后训练版本。模型采用 MoE 架构，总参数为 280 billion，每次输入激活 14 billion 参数，使用 33.5 trillion tokens 预训练，训练上下文最高为 128,000 tokens，设计上支持在微调和推理时扩展至 1 million tokens。
- Hacker News front pageBaidu released ERNIE 5.5 Flash Base
百度发布 ERNIE-5.5-Flash-Base 基础模型权重，采用稀疏 MoE 架构，总参数为 280B，每个 token 激活 14B 参数。模型从零预训练使用 33.5T tokens，训练上下文扩展至 128K，支持微调和推理时的 1M-token 上下文，发布检查点未经中期训练或后训练。
Heat over time
Heat now 16·Comparable peak 17（Nov 6 03:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.