Closed alykhantejani closed 3 months ago
Hey @alykhantejani, thank you for connecting with us. The new version should be available very soon. We noticed the dynamic_partition
issue and are working on it. We will get back to you when we get a point. Many thanks!
@rhdong ok thanks, so is still an issue in master?
Hi,
I was wondering if a new version will be cut anytime soon, I see there are many changes since a year ago.
I recently hit an error:
data.shape must start with partitions.shape, got data.shape = [2,64], partitions.shape = [1]\n\t [[{{node SGD/SGD/update_21/DynamicPartition_2}}]]
Which comes from the
DynamicEmbeddingOptimizer
trying to partition the data. I see in master you no longer usetf. dynamic_partition
and usetfra_data_flow_ops.tfra_dynamic_partition
which might solve it?
tfra_data_flow_ops.tfra_dynamic_partition is mostly same as tf.dynamic_partition except supporting additional input data type. So you may pass an illegal parameter into tf.dynamic_partition OP.
Ok so this probably isnt going to solve my issue. @rhdong you said you noticed the dynamic partition issue? Is this a known bug, perhaps you could shed more light on to what the problem is?
Closing as 0.7.0 is released
Hi,
I was wondering if a new version will be cut anytime soon, I see there are many changes since a year ago.
I recently hit an error:
Which comes from the
DynamicEmbeddingOptimizer
trying to partition the data. I see in master you no longer usetf. dynamic_partition
and usetfra_data_flow_ops.tfra_dynamic_partition
which might solve it?