[OPT/Galactica] Load large `galactica` models (#20390)

* fix `opt` bias

* revert unneeded assignment

src/transformers/models/opt/configuration_opt.py

@@ -74,6 +74,10 @@ class OPTConfig(PretrainedConfig):
             The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
         use_cache (`bool`, *optional*, defaults to `True`):
             Whether or not the model should return the last key/values attentions (not used by all models).
+        enable_bias (`bool`, *optional*, defaults to `True`):
+            Whether or not if the linear layers in the attention blocks should use the bias term.
+        layer_norm_elementwise_affine (`bool`, *optional*, defaults to `True`):
+            Whether or not if the layer norms should have learnable parameters.
 
     Example:
 
@@ -112,6 +116,8 @@ class OPTConfig(PretrainedConfig):
         pad_token_id=1,
         bos_token_id=2,
         eos_token_id=2,
+        enable_bias=True,
+        layer_norm_elementwise_affine=True,
         **kwargs
     ):
         super().__init__(
@@ -134,6 +140,9 @@ class OPTConfig(PretrainedConfig):
         self.layerdrop = layerdrop
         self.use_cache = use_cache
         self.do_layer_norm_before = do_layer_norm_before
+        # We keep these variables at `True` for backward compatibility.
+        self.enable_bias = enable_bias
+        self.layer_norm_elementwise_affine = layer_norm_elementwise_affine
 
         # Note that the only purpose of `_remove_final_layer_norm` is to keep backward compatibility
         # with checkpoints that have been fine-tuned before transformers v4.20.1

src/transformers/models/opt/modeling_opt.py

@@ -279,15 +279,18 @@ class OPTDecoderLayer(nn.Module):
             num_heads=config.num_attention_heads,
             dropout=config.attention_dropout,
             is_decoder=True,
+            bias=config.enable_bias,
         )
         self.do_layer_norm_before = config.do_layer_norm_before
         self.dropout = config.dropout
         self.activation_fn = ACT2FN[config.activation_function]
 
-        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
-        self.fc1 = nn.Linear(self.embed_dim, config.ffn_dim)
-        self.fc2 = nn.Linear(config.ffn_dim, self.embed_dim)
-        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.self_attn_layer_norm = nn.LayerNorm(
+            self.embed_dim, elementwise_affine=config.layer_norm_elementwise_affine
+        )
+        self.fc1 = nn.Linear(self.embed_dim, config.ffn_dim, bias=config.enable_bias)
+        self.fc2 = nn.Linear(config.ffn_dim, self.embed_dim, bias=config.enable_bias)
+        self.final_layer_norm = nn.LayerNorm(self.embed_dim, elementwise_affine=config.layer_norm_elementwise_affine)
 
     def forward(
         self,
@@ -507,7 +510,9 @@ class OPTDecoder(OPTPreTrainedModel):
         # with checkpoints that have been fine-tuned before transformers v4.20.1
         # see https://github.com/facebookresearch/metaseq/pull/164
         if config.do_layer_norm_before and not config._remove_final_layer_norm:
-            self.final_layer_norm = nn.LayerNorm(config.hidden_size)
+            self.final_layer_norm = nn.LayerNorm(
+                config.hidden_size, elementwise_affine=config.layer_norm_elementwise_affine
+            )
         else:
             self.final_layer_norm = None