Gabriel Rasskin comments

Results 24 comments of


                                            Gabriel Rasskin

AttributeError: 'SymbolicTensor' object has no attribute '_datatype_enum'

Thank you @innat, as a temporary workaround you can call `processing_model.build((None, 224, 224, 3))` before the loop. Alternatively, the issue also goes away by setting tensorflow to eager mode so...

Accessing dataloader and model output in callbacks

Hi @innat, would a custom metric work for this usecase? It should have access to model outputs as opposed to custom callbacks (which do not). https://keras.io/api/metrics/#creating-custom-metrics

Gradient accumulation support?

Hi @andreped, thank you for all your work on this - would you be interested in contributing a PR adding gradient accumulation as a parameter to `fit`? Multi-GPU support should...

L-BFGS-B optimizer in Keras

@fchollet @chenmoneygithub Please take a look! We're unsure whether L-BFGS-B is an optimizer we should add to Keras at this time.

steps_per_execution autotune

Hi @innat, I'm preparing to release a callback that autotunes `steps_per_execution`. We've been able to see a significant speedup especially in distributed systems. The callback is still going through testing...

steps_per_execution autotune

Hi @feelmyears, we still have to add more information on this in documentation and tutorials, but if you're using the nightly version of Keras you can try it out by...

steps_per_execution autotune

Note this is still highly experimental and final implementation/results might vary.

keras.Layers.Embedding fails with XLA and tf.distribute

Would you mind taking a look Rick? Thank you!

Keras load LSTM/GRU model with constant mask/initial_state will raise error

@nkovela1

Backward compatibility issues: arbitrary inputs and overreliance on custom layers

Hi @AlexanderLutsenko, I agree this is a pain point with Keras, working with different tensors we haven't yet standardized an indexing method. Ideally we want users to be able to...