Julian Risch

Sorry, I can't remember coming across any speed issues related to `top_k_per_sample`, no. However, I am pretty sure that there is big potential for optimizations at various points in the...

First of all, I support removing nvidia apex and adding pytorch amp. 👍 Doing quick research regarding this decision, it's what seems to be the preferred, future-proof way: https://discuss.pytorch.org/t/torch-cuda-amp-vs-nvidia-apex/74994/2 What...

> > And we would also need to ensure that the documentation explains when to use this feature, how to use this feature and what to expect from it. >...

> @julian-risch Do we happen to have any GPU resources in our CI testing so I could add a test to make sure the changes to `use_amp` work as expected?...

Hi @sjrl have you already created a PR regarding documentation? This PR here looks good to me on first sight. It's a bit unfortunate that there are so many unrelated...

Alright, thanks for the quick response. No worries about the doc string updates. It's better to have them here than not having them at all! 😄

Hi @sjrl what's the status of this PR? Would you like to get another round of reviews from @agnieszka-m and me?

@pratikkotian04 the function call that triggers the error message is `_validate_index_sync()`: https://github.com/deepset-ai/haystack/blob/325bc5466a2490ca22fedb54a4b647a3a10983c5/haystack/document_stores/faiss.py#L183 What you could when you close your application is to check whether this check is successful at that...

@agnieszka-m Thank you for the detailed feedback. All the change requests are addressed now.

@Timoeller brought up that we should include QA training and evaluation in the benchmarks as well to make sure that we do not break this core functionality with new releases.