diff --git a/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py b/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py
index 1869db9e15..6a6b2e0a0d 100644
--- a/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py
+++ b/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py
@@ -193,7 +193,7 @@ def __init__(self, prefix, config, weights):
             )
         )
         # Fuse gate and up proj
-        bias = config.mlp_bias
+        bias = getattr(config, "mlp_bias", False)
         if config.model_type == "phi3":
             self.gate_up_proj = TensorParallelColumnLinear.load_gate_up(
                 config,