Issue with Loading Sequential Models

[xchen99sdr]

tensorflow/tensorflow#70757

Reposting per tensorflow developer's recommendation.

I am encountering the same issue with pip version 2.17

Thank you

[RaulCastillo547]

I'm the original poster of the tensorflow issue. Thank you xchen99sdr for reposting it.

[james77777778]

A minimal reproducible script:

from keras import layers
from keras import models
from keras import ops
from keras import saving

CONV_WIDTH = 3
model = models.Sequential([layers.Lambda(lambda x: x[:, -CONV_WIDTH:, :])])
x = ops.ones((1, 10, 4))
y = model(x)
model.save("model.keras")
revived_model = saving.load_model("model.keras", safe_mode=False)  # <- failed

A workaround is to directly assign the value instead of using variables in layers.Lambda. I don't know why it caused the issue.

from keras import layers
from keras import models
from keras import ops
from keras import saving

model = models.Sequential([layers.Lambda(lambda x: x[:, -3:, :])])  # Not using `CONV_WIDTH`
x = ops.ones((1, 10, 4))
y = model(x)
model.save("model.keras")
revived_model = saving.load_model("model.keras", safe_mode=False)

I have submitted a PR #20076 to ultimately fix it. You can also wait for it.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No