slim dove
This model card is for the base version of Jamba. It’s a pretrained, mixture-of-experts (MoE) generative text model, with 12B active parameters and a total of 52B parameters across all experts. It supports a 256K context length, and can fit up to 140K tokens on a single 80GB GPU.
slim dove
is jamba based on mamba?
probably a mix of AI21's "Jurassic" models and the name mamba.
so yes its mamba
mild drift
Any idea if/when we will get this model to play with on OpenRouter? Looks amazing
cosmic oxide
looking into it
looks like only a base version is available (on HF) - no instruct yet, but it's coming
mild drift
are there any base versions on OpenRouter at all? I'd love to try them out for certain use cases
winter imp
are there any base versions on OpenRouter at all? I'd love to try them out for c...
If you mean base versions of any model - OR has base Mixtral and Yi, for example.
glad gazelle
Sorry to resurrect an old thread, but looks like AI21 is now publicly testing this: https://www.ai21.com/blog/announcing-jamba-instruct
