FIX Don't assume past_key_valus for encoder models

Don't assume that past_key_values is part of the model_kwargs.

This fix is similar to huggingface#2140 but for encoder-decoder models. It became
necessary after huggingface/transformers#34048
was merged into transformers.

src/peft/peft_model.py

@@ -2062,7 +2062,7 @@ def prepare_inputs_for_generation(self, *args, task_ids: torch.Tensor = None, **
         model_kwargs = self.base_model_prepare_inputs_for_generation(*args, **kwargs)
         if peft_config.peft_type == PeftType.POLY:
             model_kwargs["task_ids"] = task_ids
-        if model_kwargs["past_key_values"] is None and peft_config.peft_type == PeftType.PREFIX_TUNING:
+        if model_kwargs.get("past_key_values", None) is None and peft_config.peft_type == PeftType.PREFIX_TUNING:
             batch_size = model_kwargs["decoder_input_ids"].shape[0]
             past_key_values = self.get_prompt(batch_size)
             model_kwargs["past_key_values"] = past_key_values