It splits the DPRReaderRanker into cpu/gpu versions.
Also I noticed that the GPU test was actually failing, because the Tokenizer was generating tensors always on CPU device. With this PR it can also generate GPU based tensors. The results always have to be copied to CPU then to further process them on the CPU
This is part of https://github.com/jina-ai/executors/issues/169
It splits the DPRReaderRanker into cpu/gpu versions. Also I noticed that the GPU test was actually failing, because the Tokenizer was generating tensors always on CPU device. With this PR it can also generate GPU based tensors. The results always have to be copied to CPU then to further process them on the CPU