diff --git a/openpilot/compile.py b/openpilot/compile.py
index 9b8fca77de..0cc510cb9c 100644
--- a/openpilot/compile.py
+++ b/openpilot/compile.py
@@ -34,10 +34,16 @@ def get_random_input_tensors():
     "big_input_imgs": np.random.randn(*(1, 12, 128, 256))*256,
     "desire": np.zeros((1, 8)),
     "traffic_convention": np.array([[1., 0.]]),
-    "initial_state": np.zeros((1, 512))
+    "initial_state": np.random.randn(*(1, 512))
     #"initial_state": np.zeros((1, 768))
   }
 
+  #import pickle
+  #frames, big_frames, last_state, frame_inputs, policy_outs = pickle.load(open("openpilot/test/frame_0.pkl", "rb"))
+  #np_inputs["input_imgs"] = frames
+  #np_inputs["big_input_imgs"] = big_frames
+  #np_inputs["initial_state"] = last_state[0]
+
   #for i,k in enumerate(np_inputs.keys()):
   #  dat = open("/home/batman/openpilot/xx/ml_tools/snpe/compile_test_data/dlc_input_%d" % i, "rb").read()
   #  np_inputs[k] = np.frombuffer(dat, np.float32).reshape(np_inputs[k].shape)
@@ -172,7 +178,7 @@ def compile(input, output_fn):
     try:
       from test.test_onnx import run_onnx_torch
       torch_out = run_onnx_torch(onnx_model, np_inputs).numpy()
-      print(tinygrad_out_np, torch_out, "mse", np.sum((tinygrad_out_np-torch_out)**2))
+      print(tinygrad_out_np, torch_out, "mse", np.sum((tinygrad_out_np-torch_out)**2), "max err", np.max(np.abs((tinygrad_out_np-torch_out))))
       np.testing.assert_allclose(torch_out, tinygrad_out_np, atol=1e-4, rtol=1e-2)
     except ModuleNotFoundError:
       pass