Hi,
thank you for this great work and making it publicly available
Running cookbook/tutorials/binder_design.py against esm 3.4.0 (PyPI)
hits two API mismatches.
Environment: esm 3.4.0, transformers 4.57.6, torch 2.11.0+cu130,
Python 3.12, Tutorial file from tag v3.4.0 (827ec12).
1. _esmc → esmc
_load_hf_model (line 1236) does assert model._esmc is not None,
but EsmFold2ExperimentalModel now exposes esmc:
AttributeError: 'EsmFold2ExperimentalModel' object has no attribute '_esmc'
Affects lines 1236, 1237, 1238, 1312, 1315. Patched locally with
s/\._esmc\b/.esmc/g.
2. Removed forward kwargs
After that patch, fold_and_get_distogram (line 703) passes
calculate_confidence, disto_cond, disto_cond_mask, which
EsmFold2ExperimentalModel.forward() no longer accepts:
TypeError: EsmFold2ExperimentalModel.forward() got unexpected
keyword argument(s) ['calculate_confidence', 'disto_cond', 'disto_cond_mask']
None of the three appear anywhere in esm/models/esmfold2/experimental.py.
Is binder_design.py expected to work with 3.4.0, or should it be
pinned to an earlier esm version? I'd rather not just drop the kwargs
since disto_cond looks load-bearing for the design loop.
Thanks!
Hi,
thank you for this great work and making it publicly available
Running
cookbook/tutorials/binder_design.pyagainst esm 3.4.0 (PyPI)hits two API mismatches.
Environment: esm 3.4.0, transformers 4.57.6, torch 2.11.0+cu130,
Python 3.12, Tutorial file from tag v3.4.0 (827ec12).
1.
_esmc→esmc_load_hf_model(line 1236) doesassert model._esmc is not None,but
EsmFold2ExperimentalModelnow exposesesmc:Affects lines 1236, 1237, 1238, 1312, 1315. Patched locally with
s/\._esmc\b/.esmc/g.2. Removed forward kwargs
After that patch,
fold_and_get_distogram(line 703) passescalculate_confidence,disto_cond,disto_cond_mask, whichEsmFold2ExperimentalModel.forward()no longer accepts:None of the three appear anywhere in
esm/models/esmfold2/experimental.py.Is
binder_design.pyexpected to work with 3.4.0, or should it bepinned to an earlier esm version? I'd rather not just drop the kwargs
since
disto_condlooks load-bearing for the design loop.Thanks!