mathislucka

It is actually already there :D

Reopening this, as the device is not used for the inferencer. See: https://github.com/deepset-ai/haystack/blob/632cd1c141a8b485c6ef8695685d2d8eef3ca50f/haystack/modeling/infer.py#L229

Additionally, currently transformers does not support pytorch 1.12 (see https://github.com/huggingface/transformers/issues/17971#issuecomment-1172324921). When changing the code in inferencer to pass on the `mps` device. An error is raised during prediction: ``` Inferencing...

Also see this for the current state of covered ops for the mps backend: https://github.com/pytorch/pytorch/issues/77764

There is this dataset with cross-encoder scores for MarginMSELoss: https://huggingface.co/datasets/sentence-transformers/msmarco-hard-negatives I'd vote for implementing `MultipleNegativesRankingLoss` because MarginMSE is already used in GPL and MNRL also yields very good results. What...

Oh and training definitely makes sense. If you have labeled data, you will get much better results with training than with the out-of-the-box models.

My 2 cents: > Which sentence-transformer model(s) do we suggest for out-of-the-box use? > Now it makes sense to use and promote v5 models (?) The v5 models are only...

Yes, your understandig is correct. I understand your concerns but I think I addressed these complications: - the filter parameters can be passed as JSON representations of the parameters needed....

@sjrl already found a way to make e5 work. One thing that we could improve upon though is that e5 requires documents to be prefixed with `passage:` and queries with...

> This node uses composition with another node which is generally not something we would like for v2. Instead, we would probably prefer if the node was split in two...