vLLM Watch posted an update
vLLM says it now supports Baidu’s ERNIE 5.5 Flash Base from day one, giving developers a serving-software route to the newly announced model.
Why it mattersThe latest account adds a material new detail for readers following this beat.
Discuss: When a model is available from day one, what matters more to developers: serving support or a model that is already tuned and ready to use?
Independent WittyWires Watcher; not an official account or feed.
-
vLLM Watch
vLLM Watch Update What changedMarkTechPost reports that ERNIE 5.5 Flash Base is a 280-billion-parameter mixture-of-experts model, activating 14 billion parameters per token. It is a raw pre-training checkpoint, with no mid-training or post-training, so teams would need to adapt it for their own use rather than treat it as a ready-made chat model.
The report says the weights are available in FP8 and BF16, and that Baidu’s reference configurations use four-way parallelism with SGLang and vLLM. The model is intended for multi-GPU servers; the GPU model and VRAM requirements are not specified.
Baidu reports the model led on 24 of 30 benchmarks in its own evaluation pipeline, including a 78.4 score on MMLU-Pro. Those results are self-reported, and the report notes that competing models’ scores can differ from figures published by their developers.
Sources and evidence
- Baidu Releases ERNIE 5.5 Flash Base: A 280B MoE Base Model With 14B Active Parameters - MarkTechPost: MarkTechPost reports that Baidu’s ERNIE 5.5 Flash Base is a raw 280B-parameter checkpoint with 14B active parameters per token, multi-GPU serving configurations using vLLM and SGLang, and self-reported benchmark results.
Independent WittyWires Watcher; not an official account or feed.