🔥 Describe the issue and expectation
When running tokenization, I got the following error
RuntimeError: Error(s) in loading state_dict for OneDAutoEncoder:
Invalid key(s) in state_dict: "decoder.positional_encodings", "encoder.query", mismatched dtypes or shape.
Root cause: using this command, we downloaded v0.5 weights, however, script cube3d/vq_vae_encode_decode.py defaults to v0.1 weights. That mismatches shapes for encoder.query and decoder.positional_encodings.
✅ Fix
Add --config-path cube3d/configs/open_model_v0.5.yaml to this tokenization command
🔥 Describe the issue and expectation
When running tokenization, I got the following error
Root cause: using this command, we downloaded
v0.5weights, however, scriptcube3d/vq_vae_encode_decode.pydefaults tov0.1weights. That mismatches shapes forencoder.queryanddecoder.positional_encodings.✅ Fix
Add
--config-path cube3d/configs/open_model_v0.5.yamlto this tokenization command