Remove .item() calls

README.md

@@ -50,7 +50,7 @@ run e.g.
 
 ```bash
 python3 run_pretrained_openfold.py \
-    test.fasta \
+    target.fasta \
     data/uniref90/uniref90.fasta \
     data/mgnify/mgy_clusters_2018_12.fa \
     data/pdb70/pdb70 \

openfold/data/data_transforms.py

@@ -1103,7 +1103,7 @@ def random_crop_to_size(
     else:
         num_templates = protein["aatype"].new_zeros((1,))
 
-    num_res_crop_size = min(seq_length.item(), crop_size)
+    num_res_crop_size = min(int(seq_length), crop_size)
 
     # We want each ensemble to be cropped the same way
     g = torch.Generator(device=protein["seq_length"].device)
@@ -1111,13 +1111,13 @@ def random_crop_to_size(
         g.manual_seed(seed)
 
     def _randint(lower, upper):
-        return torch.randint(
+        return int(torch.randint(
                 lower,
                 upper + 1,
                 (1,),
                 device=protein["seq_length"].device,
                 generator=g,
-        )[0].item()
+        )[0])
 
     if subsample_templates:
         templates_crop_start = _randint(0, num_templates)

openfold/data/input_pipeline.py

@@ -156,7 +156,7 @@ def process_tensors_from_config(tensors, common_cfg, mode_cfg):
     tensors = compose(nonensembled_transform_fns(common_cfg, mode_cfg))(tensors)
 
     num_ensemble = mode_cfg.num_ensemble
-    num_recycling = tensors["no_recycling_iters"].item()
+    num_recycling = int(tensors["no_recycling_iters"])
 
     if common_cfg.resample_msa_in_recycling:
         # Separate batch per ensembling & recycling step.