Baidu releases ERNIE 5.5 Flash Base with 280 billion parameters as an open-source foundation model
Summary
- Baidu has released ERNIE 5.5 Flash Base, a 280-billion-parameter open-source foundation model using a mixture-of-experts architecture.
- The model features 14 billion active parameters per input, was pre-trained on 33.5 trillion tokens, and supports a context length up to one million tokens during inference.
- Shipped without fine-tuning or post-training, developers are expected to customize the model, with a post-trained version planned for later.
- Model weights are available on Hugging Face in FP8 and BF16 formats.
Why it matters
The release of ERNIE 5.5 Flash Base gives developers access to a large-scale mixture-of-experts open foundation model with massive context capabilities, enabling custom fine-tuning and architectural experimentation.