fix: force correct dtype in HF load

modeling_hyena.py

@@ -46,6 +46,9 @@ class StripedHyenaModelForCausalLM(StripedHyenaPreTrainedModel):
         self.vocab_size = vocab_size
         self.post_init()
 
+    def post_init(self):
+        self.backbone.to_bfloat16_except_poles_residues() 
+        
     def _set_gradient_checkpointing(self, enable, gradient_checkpointing_func):
         self.backbone.gradient_checkpointing = enable