Is the architecture for this model different from the 7b variant?
#7
by
tsriharsha
- opened
Should the architecture for this model be the same as the 7b in the config.json?
LlavaForConditionalGeneration?
Hi
@tsriharsha
yes they are the same arch, it should be already the case for this model I think no?
In the 7b model, the architecture is specified as LlavaForConditionalGeneration in the config.json; but in the 13b model, the architecture is LlavaForCausalLM.
Any update on this?
Yes that's an artifact, during the addition of LLaVa to the Transformers library the model was first pushed to the hub using LlavaForCausalLM but was then renamed to LlavaForConditionalGeneration. Will update the config. Thanks for reporting