Bug description
When I try to fold the sequence of Q49HI2 (seq.txt) on a A800 GPU (80GB) by running:
python esmfold.py -i ./Q49HI2.fa -o ./ --cpu-offload
The log is "23/12/11 22:57:36 | INFO | root | Predicted structure for Q49HI2 with length 1058, pLDDT 88.0, pTM 0.999 in 139.9s. 1 / 1 completed."
Note that the pTM is 0.999, which is obviously unreasonable. And the output structure is also really weird:
The pdb file is attached here (I change the suffix to .txt for the convenience to upload):
Q49HI2.txt
Interestingly, when I set the --chunk-size parameter, everything goes back to normal:
python esmfold.py -i ./Q49HI2.fa -o ./ --cpu-offload --chunk-size 128
The log is "23/12/11 22:49:10 | INFO | root | Predicted structure for Q49HI2 with length 1058, pLDDT 65.8, pTM 0.605 in 143.8s. 1 / 1 completed."
Bug description When I try to fold the sequence of Q49HI2 (seq.txt) on a A800 GPU (80GB) by running:
python esmfold.py -i ./Q49HI2.fa -o ./ --cpu-offload
The log is "23/12/11 22:57:36 | INFO | root | Predicted structure for Q49HI2 with length 1058, pLDDT 88.0, pTM 0.999 in 139.9s. 1 / 1 completed." Note that the pTM is 0.999, which is obviously unreasonable. And the output structure is also really weird: The pdb file is attached here (I change the suffix to .txt for the convenience to upload): Q49HI2.txtInterestingly, when I set the --chunk-size parameter, everything goes back to normal:
python esmfold.py -i ./Q49HI2.fa -o ./ --cpu-offload --chunk-size 128
The log is "23/12/11 22:49:10 | INFO | root | Predicted structure for Q49HI2 with length 1058, pLDDT 65.8, pTM 0.605 in 143.8s. 1 / 1 completed."