fix typos in STG attention

custom_cogvideox_transformer_3d.py

@@ -130,8 +130,8 @@ class CogVideoXAttnProcessor2_0:
             hidden_states_uncond, hidden_states_cond, hidden_states_perturb = hidden_states.chunk(3, dim=0)
             encoder_hidden_states_uncond, encoder_hidden_states_cond, encoder_hidden_states_perturb = encoder_hidden_states.chunk(3, dim=0)
 
-            hidden_states = torch.cat([hidden_states_uncond, hidden_states], dim=0)
-            encoder_hidden_states = torch.cat([encoder_hidden_states_uncond, encoder_hidden_states], dim=0)
+            hidden_states = torch.cat([hidden_states_uncond, hidden_states_cond], dim=0)
+            encoder_hidden_states = torch.cat([encoder_hidden_states_uncond, encoder_hidden_states_cond], dim=0)
 
         text_seq_length = encoder_hidden_states.size(1)
 
@@ -197,27 +197,27 @@ class CogVideoXAttnProcessor2_0:
             hidden_states *= feta_scores
 
         if self.stg_mode == "STG-A":
-            text_seq_length = encoder_hidden_states.size(1)
+            text_seq_length = encoder_hidden_states_perturb.size(1)
 
-            hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
+            hidden_states_perturb = torch.cat([encoder_hidden_states_perturb, hidden_states_perturb], dim=1)
 
             batch_size, sequence_length, _ = (
-                hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+                hidden_states_perturb.shape if encoder_hidden_states_perturb is None else encoder_hidden_states_perturb.shape
             )
 
-            if attention_mask is not None:
-                attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
-                attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+            #if attention_mask is not None:
+            #    attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            #    attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
 
             if attn.to_q.weight.dtype == torch.float16 or attn.to_q.weight.dtype == torch.bfloat16:
-                hidden_states = hidden_states.to(attn.to_q.weight.dtype)
+                hidden_states_perturb = hidden_states_perturb.to(attn.to_q.weight.dtype)
 
             if not "fused" in self.attention_mode:
-                query_perturb = attn.to_q(hidden_states)
-                key_perturb = attn.to_k(hidden_states)
-                value_perturb = attn.to_v(hidden_states)
+                query_perturb = attn.to_q(hidden_states_perturb)
+                key_perturb = attn.to_k(hidden_states_perturb)
+                value_perturb = attn.to_v(hidden_states_perturb)
             else:
-                qkv = attn.to_qkv(hidden_states)
+                qkv = attn.to_qkv(hidden_states_perturb)
                 split_size = qkv.shape[-1] // 3
                 query_perturb, key_perturb, value_perturb = torch.split(qkv, split_size, dim=-1)
 
@@ -254,19 +254,19 @@ class CogVideoXAttnProcessor2_0:
             )
 
             if self.attention_mode != "comfy":
-                hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+                hidden_states_perturb = hidden_states_perturb.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
 
             # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
+            hidden_states_perturb = attn.to_out[0](hidden_states_perturb)
             # dropout
-            hidden_states = attn.to_out[1](hidden_states)
+            hidden_states_perturb = attn.to_out[1](hidden_states_perturb)
 
-            encoder_hidden_states, hidden_states = hidden_states.split(
+            encoder_hidden_states_perturb, hidden_states_perturb = hidden_states_perturb.split(
                 [text_seq_length, hidden_states.size(1) - text_seq_length], dim=1
             )
 
-            hidden_states = torch.cat([hidden_states_org, hidden_states_perturb], dim=0)
-            encoder_hidden_states = torch.cat([encoder_hidden_states_org, encoder_hidden_states_perturb], dim=0)
+            hidden_states = torch.cat([hidden_states, hidden_states_perturb], dim=0)
+            encoder_hidden_states = torch.cat([encoder_hidden_states, encoder_hidden_states_perturb], dim=0)
 
         return hidden_states, encoder_hidden_states