Support GPT-NeoX Models without attention biases (#2301)

Author: dalgarak

vllm/model_executor/models/gpt_neox.py

@@ -54,6 +54,7 @@ def __init__(
         self.total_num_heads = config.num_attention_heads
         self.hidden_size = config.hidden_size
         self.head_size = self.hidden_size // self.total_num_heads
+        self.bias = getattr(config, "attention_bias", True)
 
         tensor_model_parallel_world_size = (
             get_tensor_model_parallel_world_size())
@@ -65,11 +66,13 @@ def __init__(
             config.hidden_size,
             self.head_size,
             self.total_num_heads,
+            bias=self.bias,
             linear_method=linear_method,
         )
         self.dense = RowParallelLinear(
             config.hidden_size,
             config.hidden_size,
+            bias=self.bias,
             linear_method=linear_method,
         )
         scaling = self.head_size**-0.5