Hello. How long real max context for this model? Config says that it's same as mixtral/mistral - 32k. But in reality it produces just noise even on 20k (while original mixtral handle it correctly).
Hi @Desm0nt this is a model made with 4 Mistral-7b models with a context of 8k and not a Mixtral model with a context of 32k. This behavior is expected.
I updated the model and the config.json to reflect that. I hope it's clearer now.