Hi,
Thanks to your amazing team for open-sourcing the ESM series. It's been a huge help to the research community!
Could you please validate the correctness of the backbone coordinate conversion scripts in this fork: https://github.com/mahdip72/esm
Specifically, these two:
These scripts are being used for benchmarking ESM3's structure tokenization performance against other methods. We want to ensure the coordinate-to-token-to-coordinate roundtrip is accurate.
The scripts use ESM3_structure_encoder_v0 for encoding and ESM3_structure_decoder_v0 for decoding with mixed precision (FP16) and batch processing.
Any feedback or fixes would be super helpful to keep our benchmarks accurate!
Thanks a lot!
Hi,
Thanks to your amazing team for open-sourcing the ESM series. It's been a huge help to the research community!
Could you please validate the correctness of the backbone coordinate conversion scripts in this fork: https://github.com/mahdip72/esm
Specifically, these two:
These scripts are being used for benchmarking ESM3's structure tokenization performance against other methods. We want to ensure the coordinate-to-token-to-coordinate roundtrip is accurate.
The scripts use ESM3_structure_encoder_v0 for encoding and ESM3_structure_decoder_v0 for decoding with mixed precision (FP16) and batch processing.
Any feedback or fixes would be super helpful to keep our benchmarks accurate!
Thanks a lot!