From 3adc67c7a498f3a38132658282875f1b95ffed1c Mon Sep 17 00:00:00 2001
From: somebody <onesome01@protonmail.com>
Date: Mon, 13 Mar 2023 19:37:45 -0500
Subject: [PATCH] RWKV: Move import right before usage

So we don't needlessly compile the cuda kernel
---
 aiserver.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aiserver.py b/aiserver.py
index 6ebb4b8b..b1a16447 100644
--- a/aiserver.py
+++ b/aiserver.py
@@ -592,7 +592,6 @@ from modeling.inference_models.legacy_gpt2_hf import CustomGPT2HFTorchInferenceM
 from modeling.inference_models.hf_mtj import HFMTJInferenceModel
 from modeling.inference_models.horde import HordeInferenceModel
 from modeling.inference_models.openai import OpenAIAPIInferenceModel
-from modeling.inference_models.rwkv import RWKVInferenceModel
 
 
 old_socketio_on = socketio.on
@@ -1894,6 +1893,7 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
     elif "rwkv" in koboldai_vars.model:
         if koboldai_vars.use_colab_tpu:
             raise RuntimeError("RWKV is not supported on the TPU.")
+        from modeling.inference_models.rwkv import RWKVInferenceModel
         model = RWKVInferenceModel(koboldai_vars.model)
         model.load()
     elif not koboldai_vars.use_colab_tpu and not koboldai_vars.noai: