Pass LoRA rank to LoRALinearLayer (huggingface#2191)

Author: asadm

models/cross_attention.py

@@ -296,10 +296,10 @@ class LoRACrossAttnProcessor(nn.Module):
     def __init__(self, hidden_size, cross_attention_dim=None, rank=4):
         super().__init__()
 
-        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size)
-        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size)
-        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size)
-        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size)
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank)
 
     def __call__(
         self, attn: CrossAttention, hidden_states, encoder_hidden_states=None, attention_mask=None, scale=1.0
@@ -408,10 +408,10 @@ class LoRAXFormersCrossAttnProcessor(nn.Module):
     def __init__(self, hidden_size, cross_attention_dim, rank=4):
         super().__init__()
 
-        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size)
-        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size)
-        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size)
-        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size)
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank)
 
     def __call__(
         self, attn: CrossAttention, hidden_states, encoder_hidden_states=None, attention_mask=None, scale=1.0