Krzysztof Lecki comments

Results 66 comments of


                                            Krzysztof Lecki

fix: fix a typo

!build

Wrong accuracy with external data in a distributed setting

Hi @thomas-bouvier, my suggestion would be to check the behaviour around the end of the epoch. By default DALIGenericIterator uses [Fill policy](https://docs.nvidia.com/deeplearning/dali/main-user-guide/docs/advanced_topics_sharding.html#nvidia.dali.plugin.base_iterator.LastBatchPolicy.FILL), and the reader you used is configured as...

Wrong accuracy with external data in a distributed setting

Hi, sorry, I linked the wrong page for the debug mode: https://docs.nvidia.com/deeplearning/dali/user-guide/docs/examples/general/debug_mode.html Can you elaborate on the: > First thing I observed though: the first batch doesn't contain the same...

Wrong accuracy with external data in a distributed setting

You can also check the validation with external source against the network trained with file reader (or with other preprocessing) and other way round. That could probably help to narrow...

Can we have OneOf operator in Dali?

Hi @twmht we are currently working on introducing support for conditional execution in DALI. In the current form, you would need to compute all the possible options eagerly, and than...

[WiP] TensorVector Rework - for CI runs

!build

[WiP] TensorVector Rework - for CI runs

Changes were merged as a separate PR

[RFC] Use TensorVector everywhere

!build

[RFC] Use TensorVector everywhere

Changes were merged as a separate PRs

Is it possible to execute custom operation on multiple cuda streams?

Hi @baneling100, DALI Pipeline executor uses one stream for GPU-stage (all GPU operators). It runs the operators in topological order. In theory you could create 13 streams in the instance...