Update llama_xformers_attention.py
Browse files
llama_xformers_attention.py
CHANGED
@@ -70,10 +70,6 @@ class LlamaXFormersAttention(LlamaAttention):
|
|
70 |
attn_weights = None
|
71 |
|
72 |
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
|
73 |
-
|
74 |
attn_output = self.o_proj(attn_output)
|
75 |
|
76 |
-
if not output_attentions:
|
77 |
-
attn_weights = None
|
78 |
-
|
79 |
return attn_output, attn_weights, past_key_value
|
|
|
70 |
attn_weights = None
|
71 |
|
72 |
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
|
|
|
73 |
attn_output = self.o_proj(attn_output)
|
74 |
|
|
|
|
|
|
|
75 |
return attn_output, attn_weights, past_key_value
|