Cannot export model for inference (via saved_model format)

[Andrei997]

Hi, and first of all thank you for your contribution. To cut it short, I cannot save a trained model using the saved_model format, with the following error: TypeError: Input 'y' of 'AddV2' Op has type bool that does not match type float32 of argument 'x'

The code I'm using to save a model is: model.save("./exports", include_optimizer=False)

I need to emphasize that I do not want to save the model via save_weights, as I need to use the saved_model format for inference with tools like TensorflowServing and TritonInferenceServer.

Any help would be much appreciated!

Problem Cannot save model via model.save(...) or tf.saved_model.save(...)

Minimal steps to reproduce issue

from detr_tf.networks.detr import get_detr_model
from detr_tf.optimizers import setup_optimizers
from detr_tf import training
from detr_tf.inference import get_model_inference, numpy_bbox_to_image

detr = get_detr_model(config, include_top=False, nb_class=2, weights="detr")
detr.save("./exports",  include_optimizer=False)

[Andrei997]

Updated, changing attn_output_weights += attn_mask to attn_output_weights += tf.cast(attn_mask, tf.float32) in transformer.py 320 fixes the initial problem, but another problem appears when exporting the model.

The new error is: ValueError: None values not supported., and it seems to come from: query_tgt = key_tgt = target + query_encoding in transformer.py 211 ... Any ideas ?

[Andrei997]

Update: seems it's possible to save the model to a .pb file (frozen graph), though I haven't tried to load the model and use it for inference yet.

For reference, I managed to save the model with:

g = tf.Graph()

with g.as_default():

    model = get_detr_model(config, include_top=False, nb_class=2, weights="detr")

    input = tf.random.uniform((1, 512, 512, 3), minval=0, maxval=1, dtype=tf.dtypes.float32)

    output = model(input)

    tf.io.write_graph(g.as_graph_def(), './exports', 'model.pb', as_text=False)

It would still be nice to be able to save the model via the recommended way, using model.save(...), so any suggestions regarding that would be appreciated.

[ni3s]

I follow below approach to save model and its checkpoint and to load it again

os.makedirs("/data/W2FORMS/",exist_ok=True) checkpoint_path = "/data/W2FORMS/detr.ckpt" checkpoint_dir = os.path.dirname(checkpoint_path)

for epoch in range(0,600 ): training.fit(detr, dataset, optimzers, config, epoch_nb=epoch, class_names=CLASS_NAMES) detr.save_weights(checkpoint_path)

Now for loading this weights again to the model

• load the base default model ( model = get_detr_model(config, include_top=False, nb_class=2, weights="detr"))
• Do all the configuration we did before we use training.fit
• then use (detr.load_weights(filepath='/data/W2FORMS/detr.ckpt',by_name=False)) to load above saved weights