diff --git a/server/text_generation_server/models/custom_modeling/flash_cohere_modeling.py b/server/text_generation_server/models/custom_modeling/flash_cohere_modeling.py
index 56d9a966429..1d4620c71be 100644
--- a/server/text_generation_server/models/custom_modeling/flash_cohere_modeling.py
+++ b/server/text_generation_server/models/custom_modeling/flash_cohere_modeling.py
@@ -216,7 +216,9 @@ def __init__(
 
         self.query_key_value = load_attention(config, prefix, weights)
 
-        self.use_qk_norm = config.use_qk_norm
+        self.use_qk_norm = (
+            config.use_qk_norm if hasattr(config, "use_qk_norm") else False
+        )
         if self.use_qk_norm:
             self.q_norm = CohereLayerNorm(
                 prefix=f"{prefix}.q_norm",