Keypoint Detection
Transformers
Safetensors
vitpose

MOE

#1
by Neman - opened

This is MOE model, yes?
ValueError: dataset_index must be provided when using multiple experts (num_experts=6). Please provide dataset_index to the forward pass.

So it needed to change model call to:
with torch.no_grad():
outputs = model(**inputs, dataset_index=torch.tensor([0]).to(device)) #first expert

What are the differences between experts; is there a list for what each is trained for? People, animals? Robots? :)

University of Sydney org

Hi,

Yes it's an MoE model which uses 6 experts. This is explained here: https://huggingface.co/docs/transformers/en/model_doc/vitpose#vitpose-models

Thank you for info.
Maybe you could also correct the code in the model card.

Neman changed discussion status to closed

Sign up or log in to comment