fix: add adapter_data param to phi and neox

server/text_generation_server/models/custom_modeling/flash_neox_modeling.py

@@ -390,6 +390,7 @@ class FlashGPTNeoXForCausalLM(FlashGPTNeoXPreTrainedModel):
         max_s: int,
         prefill_cache_indices: Optional[torch.Tensor],
         lm_head_indices: Optional[torch.Tensor] = None,
+        adapter_data: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
         hidden_states = self.gpt_neox(
             input_ids,

server/text_generation_server/models/custom_modeling/flash_phi_modeling.py

@@ -400,6 +400,7 @@ class FlashPhiForCausalLM(torch.nn.Module):
         max_s: int,
         prefill_cache_indices: Optional[torch.Tensor],
         lm_head_indices: Optional[torch.Tensor] = None,
+        adapter_data: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
         hidden_states = self.model(
             input_ids,