Internal error in tflite.Interpreter

Ankit Bharti

Getting this error in example custom pretrained model mobilenet_v1_1.0_224.tflite from model zoo.

Code

# create the output dir if not preset
# clear the directory
os.makedirs(output_dir, exist_ok=True)
for root, dirs, files in os.walk(output_dir, topdown=False):
    [os.remove(os.path.join(root, f)) for f in files]
    [os.rmdir(os.path.join(root, d)) for d in dirs]

tidl_delegate = [tflite.load_delegate(os.path.join(os.environ['TIDL_TOOLS_PATH'], 'tidl_model_import_tflite.so'), compile_options)]
interpreter = tflite.Interpreter(model_path=tflite_model_path, experimental_delegates=tidl_delegate)
interpreter.allocate_tensors()

input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()

for num in tqdm.trange(len(calib_images)):
    interpreter.set_tensor(input_details[0]['index'], preprocess_for_tflite_mobilenetv1(calib_images[num]))
    interpreter.invoke()

over 2 years ago

0 Anand Pathak over 2 years ago

TI__Genius 9065 points

Can you check if model works fine without TIDL offload (Remove experimental_delegates=tidl_delegate from the interpreter) and check if model works fine? Are you using Cloud to try out the model?

Regards,

Anand

0 Yongquan Xie over 2 years ago in reply to Anand Pathak

Prodigy 50 points

I'm getting exactly the same.

Then I tried: interpreter = tflite.Interpreter(model_path=tflite_model_path), got same error.

The trick is, I run exactly the same using the same tflite about a few weeks ago. it worked well at that time. The notebook is saved in the cloud space I did not change anything, now it failed.

Any suggestions?

0 Paula Carrillo over 2 years ago in reply to Yongquan Xie

TI__Mastermind 39260 points

Hi Yongquan, just to confirm, are you running your notebook in the cloud? if so, as mentioned in another thread we have some instability issues. If you are running on your host then we will need more details. Ex. PSDK version you are using, model type or an example to reproduce it.

thank you,

Paula