fixes

codes/models/clip/clvp.py

@@ -92,7 +92,7 @@ class CLVP(nn.Module):
         return {
             'conditioning': list(self.conditioning_transformer.parameters()),
             'text': list(self.text_transformer.parameters()),
-            'speech': list(self.speech_transformer.parameters()) + list(self.mel_head.parameters()),
+            'speech': list(self.speech_transformer.parameters()),
         }
 
     def forward(

codes/models/lucidrains/x_transformers.py

@@ -779,6 +779,7 @@ class AttentionLayers(nn.Module):
         self.dim = dim
         self.depth = depth
         self.layers = nn.ModuleList([])
+        self.causal = causal
 
         rel_pos_bias = 'rel_pos_bias' in attn_kwargs
         self.has_pos_emb = position_infused_attn or rel_pos_bias or rotary_pos_emb
@@ -921,6 +922,7 @@ class AttentionLayers(nn.Module):
             return_hiddens=False,
             norm_scale_shift_inp=None,
             past_key_values=None,
+            expected_seq_len=None,
     ):
 
         assert not (self.cross_attend ^ (exists(context) or exists(
@@ -939,10 +941,14 @@ class AttentionLayers(nn.Module):
 
         rotary_pos_emb = None
         if exists(self.rotary_pos_emb):
+            if not self.training and self.causal:
+                assert expected_seq_len is not None, "To decode a transformer with rotary embeddings, you must specify an `expected_seq_len`"
+            elif expected_seq_len is None:
+                expected_seq_len = 0
             seq_len = x.shape[1]
             if past_key_values is not None:
                 seq_len += past_key_values[0][0].shape[-2]
-            max_rotary_emb_length = max(list(map(lambda m: (m.shape[1] if exists(m) else 0) + seq_len, mems)))
+            max_rotary_emb_length = max(list(map(lambda m: (m.shape[1] if exists(m) else 0) + seq_len, mems)) + [expected_seq_len])
             rotary_pos_emb = self.rotary_pos_emb(max_rotary_emb_length, x.device)
 
         present_key_values = []