fairseq/modules/transformer_layer.py

Survived

Mutant 715

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -158,7 +158,7 @@
     """
 
     def __init__(
-        self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=False
+        self, args, no_encoder_attn=True, add_bias_kv=False, add_zero_attn=False
     ):
         super().__init__()
         self.embed_dim = args.decoder_embed_dim

Mutant 716

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -158,7 +158,7 @@
     """
 
     def __init__(
-        self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=False
+        self, args, no_encoder_attn=False, add_bias_kv=True, add_zero_attn=False
     ):
         super().__init__()
         self.embed_dim = args.decoder_embed_dim

Mutant 717

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -158,7 +158,7 @@
     """
 
     def __init__(
-        self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=False
+        self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=True
     ):
         super().__init__()
         self.embed_dim = args.decoder_embed_dim

Mutant 718

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -216,7 +216,7 @@
     def build_fc2(self, input_dim, output_dim, q_noise, qn_block_size):
         return quant_noise(nn.Linear(input_dim, output_dim), q_noise, qn_block_size)
 
-    def build_self_attention(self, embed_dim, args, add_bias_kv=False, add_zero_attn=False):
+    def build_self_attention(self, embed_dim, args, add_bias_kv=True, add_zero_attn=False):
         return MultiheadAttention(
             embed_dim,
             args.decoder_attention_heads,

Mutant 719

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -216,7 +216,7 @@
     def build_fc2(self, input_dim, output_dim, q_noise, qn_block_size):
         return quant_noise(nn.Linear(input_dim, output_dim), q_noise, qn_block_size)
 
-    def build_self_attention(self, embed_dim, args, add_bias_kv=False, add_zero_attn=False):
+    def build_self_attention(self, embed_dim, args, add_bias_kv=False, add_zero_attn=True):
         return MultiheadAttention(
             embed_dim,
             args.decoder_attention_heads,

Mutant 720

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -254,7 +254,7 @@
         self_attn_mask: Optional[torch.Tensor] = None,
         self_attn_padding_mask: Optional[torch.Tensor] = None,
         need_attn: bool = False,
-        need_head_weights: bool = False,
+        need_head_weights: bool = True,
     ):
         """
         Args:

Mutant 721

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -379,7 +379,7 @@
             return x, attn, self_attn_state
         return x, attn, None
 
-    def make_generation_fast_(self, need_attn: bool = False, **kwargs):
+    def make_generation_fast_(self, need_attn: bool = True, **kwargs):
         self.need_attn = need_attn
 
 

Mutant 722

--- fairseq/modules/transformer_layer.py
+++ fairseq/modules/transformer_layer.py
@@ -383,7 +383,7 @@
         self.need_attn = need_attn
 
 
-def Linear(in_features, out_features, bias=True):
+def Linear(in_features, out_features, bias=False):
     m = nn.Linear(in_features, out_features, bias)
     nn.init.xavier_uniform_(m.weight)
     if bias: