Hi,
Thank you for providing this dataset.
I noticed a small point of confusion regarding the Instant-NGP checkpoints. In the README, you mention that the models were trained using a fork that relies on CUTLASS MLP, but when inspecting the .msgpack files, both network and rgb_network appear to have otype: FullyFusedMLP.
Could you clarify which implementation was actually used during training?
Hi,
Thank you for providing this dataset.
I noticed a small point of confusion regarding the Instant-NGP checkpoints. In the README, you mention that the models were trained using a fork that relies on CUTLASS MLP, but when inspecting the .msgpack files, both network and rgb_network appear to have otype: FullyFusedMLP.
Could you clarify which implementation was actually used during training?