diff --git a/server/text_generation_server/models/custom_modeling/flash_mistral_modeling.py b/server/text_generation_server/models/custom_modeling/flash_mistral_modeling.py
index a11ab1d3..eca01bbb 100644
--- a/server/text_generation_server/models/custom_modeling/flash_mistral_modeling.py
+++ b/server/text_generation_server/models/custom_modeling/flash_mistral_modeling.py
@@ -120,7 +120,7 @@ class MistralAttention(torch.nn.Module):
         if hasattr(config, "head_dim"):
             self.head_size = config.head_dim
         else:
-            self.head_size = hidden_size // num_attention_heads
+            self.head_size = self.hidden_size // self.num_heads
 
         self.rotary_emb = PositionRotaryEmbedding.static(
             config=config,