metadata
license: apache-2.0
Here is a inference demo on Colab.
Please note that openmoe-base is only a small, debugging-focused model and hasn't been extensively trained. Oftentimes, its output is not valid.
If you want to see a decent performance on QAs, please try the openmoe-8b(with compute comparable to a 1.6B LLaMA), or the openmoe-34B version later.