qkv

fa3.py

@@ -36,9 +36,9 @@ class FlashFusedFluxAttnProcessor3_0:
         batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
 
         # `sample` projections.
-        qkv = attn.to_qkv(hidden_states)
-        split_size = qkv.shape[-1] // 3
-        query, key, value = torch.split(qkv, split_size, dim=-1)
+        query = attn.to_q(hidden_states)
+        key = attn.to_k(hidden_states)
+        value = attn.to_v(hidden_states)
 
         inner_dim = key.shape[-1]
         head_dim = inner_dim // attn.heads