Can't deploy kmodel model on K210 (firmware 0.6.2, nncase 1.1.0)

Describe the bug I am trying to deploy a custom model on Maix Dock I, which firmware is 0.6.2. The board crashed when I try to load the model.

To Reproduce

The model I try to deploy is the following:

# create network
import tensorflow as tf

inputs = tf.keras.layers.Input((28,28,1))
x = tf.keras.layers.Conv2D(32,5,activation="relu")(inputs)
f = tf.keras.layers.Flatten()(x)
out = tf.keras.layers.Dense(10, activation="softmax")(f)
model = tf.keras.models.Model(inputs=inputs, outputs=out)
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3), loss="categorical_crossentropy")

Then I convert it to ONNX using tf2onnx

import tf2onnx
import onnxruntime as rt

spec = (tf.TensorSpec((None, 28, 28, 1), tf.float32, name="input"),)
output_path = model.name + ".onnx"

model_proto, _ = tf2onnx.convert.from_keras(model, input_signature=spec, opset=13, output_path=output_path)

Finally, I convert it to kmodel using the tutorial found in the doc:

model_file = 'model.onnx'
target = 'k210'

# onnx simplify from documentation
model_file = onnx_simplify(model_file)

# compile_options
compile_options = nncase.CompileOptions()
compile_options.target = target
compile_options.dump_ir = True
compile_options.dump_asm = True
compile_options.dump_dir = 'tmp'
# quantize model
# compile_options.quant_type = 'uint8' # or 'int8'
# compiler
compiler = nncase.Compiler(compile_options)

# import_options
import_options = nncase.ImportOptions()

# import
model_content = read_model_file(model_file)
compiler.import_onnx(model_content, import_options)

# compile
compiler.compile()

# kmodel
kmodel = compiler.gencode_tobytes()
with open('test.kmodel', 'wb') as f:
    f.write(kmodel)

I then upload the model to /flash/ of the Maix Dock using rshell. However, when I try to load it in micropython

...
kpu.load("/flash/test.kmodel")

The board crashes without any error nor debug message.

What could be the source of the problem?

Expected behavior

The model should be loaded in the KPU and allow inference, or an error message should be displayed.

Origin model and code

This zip contains the different models generated by the code. models.zip

Environment (please complete the following information):

OS: MacOS Catalina
Python 3.7.9
TF: 2.5.0
ONNX: 1.10.2
tf2onnx: 1.9.2
ONNX-simplify: 0.3.6
nncase: nncase-1.1.0.20211203

kendryte / nncase

Can't deploy kmodel model on K210 (firmware 0.6.2, nncase 1.1.0) #462