Update VERSION to 1.20.0dev

Signed-off-by: Kamil Tokarski <[email protected]>

Fix TL0_python-self-test-base-cuda CUDA setup (NVIDIA#4362)

- fixes the changes and restoration of the previous cuda env setup
  in TL0_python-self-test-base-cuda test

Signed-off-by: Janusz Lisiecki <[email protected]>

Enable CUDA 11.8 and Hopper support (NVIDIA#4308)

- adds CUDA 11.8 based build
- extend the image decoder to support H100 Hopper

Signed-off-by: Janusz Lisiecki <[email protected]>

Enable CUDA 11.8 and Hopper support (NVIDIA#4308)

- adds CUDA 11.8 based build
- extend the image decoder to support H100 Hopper

Signed-off-by: Janusz Lisiecki <[email protected]>

Update VERSION to 1.19.0dev

Signed-off-by: Kamil Tokarski <[email protected]>

Update VERSION to 1.18.0dev

Signed-off-by: Kamil Tokarski <[email protected]>

Fix device pick in Numpy reader tests (NVIDIA#4104)

Signed-off-by: Albert Wolant <[email protected]>

Fix lack of reset when `iter()` is called on the DALI framework itera…

…tor (NVIDIA#4048)

- the DALI framework iterator when `iter()` is called checks if any
  data has been consumed yet. If it hasn't it doesn't reset to prevent
  improper operation when DALI FW iterator prefetches the first batch after
  creation, and invocation of something like `enumerate(iterator)`,
  would reset the iterator and extend the length of the first epoch
  (prefetched data in the FW iterator would be added to the freshly
  reset iterator). The flag that is set when any data has been consumed
  was set only in the `next()` method and not `__next__()`. This PR
  fixes this problem and adjusts the test

Signed-off-by: Janusz Lisiecki <[email protected]>

Update VERSION to 1.17.0dev

Signed-off-by: Kamil Tokarski <[email protected]>

Fix dltensor operator tests (NVIDIA#3984)

* Fix dltensor operator tests (missing imports)
* Fix backend_impl shadowing Pipeline import

Signed-off-by: Kamil Tokarski <[email protected]>