--- /usr/local/lib/python3.7/dist-packages/alphafold/model/modules.py	2022-03-23 04:02:56.871333044 +0000
+++ alphafold/alphafold/model/modules.py	2022-03-23 04:04:16.081234108 +0000
@@ -341,17 +341,16 @@
           compute_loss=compute_loss,
           ensemble_representations=ensemble_representations)
 
+    emb_config = self.config.embeddings_and_evoformer
+    prev = {
+        'prev_pos': jnp.zeros(
+            [num_residues, residue_constants.atom_type_num, 3]),
+        'prev_msa_first_row': jnp.zeros(
+            [num_residues, emb_config.msa_channel]),
+        'prev_pair': jnp.zeros(
+            [num_residues, num_residues, emb_config.pair_channel]),
+    }
     if self.config.num_recycle:
-      emb_config = self.config.embeddings_and_evoformer
-      prev = {
-          'prev_pos': jnp.zeros(
-              [num_residues, residue_constants.atom_type_num, 3]),
-          'prev_msa_first_row': jnp.zeros(
-              [num_residues, emb_config.msa_channel]),
-          'prev_pair': jnp.zeros(
-              [num_residues, num_residues, emb_config.pair_channel]),
-      }
-
       if 'num_iter_recycling' in batch:
         # Training time: num_iter_recycling is in batch.
         # The value for each ensemble batch is the same, so arbitrarily taking
@@ -378,7 +377,6 @@
             body,
             (0, prev))
     else:
-      prev = {}
       num_iter = 0
 
     ret = do_call(prev=prev, recycle_idx=num_iter)