invalid shape type=16

[patrickbryant1]

Hi, I am trying to predict a complex with stoichiometry A6B6 using AFM v2.3. I get a strange error related to something in tf, but don't really know why. Using the same MSAs, I can predict a smaller version of the complex so this seems to be related to the number of chains. Is there an issue with larger complexes?

NVIDIA A100 GPU.

Running model model_1_multimer_v3_pred_0 I0515 09:27:27.215354 47755502927936 model.py:166] Running predict with shape(feat) = {'aatype': (5328,), 'residue_index': (5328,), 'seq_length': (), 'msa': (3208, 5328), 'num_alignments': (), 'template_aatype': (4, 5328), 'template_all_atom_mask': (4, 5328, 37), 'template_all_atom_positions': (4, 5328, 37, 3), 'asym_id': (5328,), 'sym_id': (5328,), 'entity_id': (5328,), 'deletion_matrix': (3208, 5328), 'deletion_mean': (5328,), 'all_atom_mask': (5328, 37), 'all_atom_positions': (5328, 37, 3), 'assembly_num_chains': (), 'entity_mask': (5328,), 'num_templates': (), 'cluster_bias_mask': (3208,), 'bert_mask': (3208, 5328), 'seq_mask': (5328,), 'msa_mask': (3208, 5328)} 2023-05-15 09:28:23.786370: E external/org_tensorflow/tensorflow/compiler/xla/shape_util.cc:311] INVALID_ARGUMENT: invalid shape type=16, dims=[-661356544] 2023-05-15 09:28:23.787370: F external/org_tensorflow/tensorflow/tsl/platform/statusor.cc:33] Attempting to fetch value instead of handling error INVALID_ARGUMENT: invalid shape type=16, dims=[-661356544] Fatal Python error: Aborted

[dinahoch]

Hi, did you ever find a solution to this problem?

[patrickbryant1]

No