Matthijs Douze
Matthijs Douze
About the merge_into inconsistency: this is because the baseline_index (that manages index.ivfdata) is opened twice, once as baseline_index and once as empty_trained in add_data. Therefore there is an inconsistency between...
this may be because both Faiss and pymupdf are wrapped with SWIG. LMC if there is a workaround for this case.
I think we could use SWIG_TYPE_TABLE to make a unique type table for Faiss. https://www.swig.org/Doc4.2/Modules.html#Modules_nn2 It seems that it just makes sure the table holding type names is distinct for...
``` m = list(set(self.pool.map(lambda idx: idx.metric_type, self.sub_indexes))) ``` shouldn't it call idx.get_metric_type() ?
right, adding this as enhancement.
Would it be possible to simplify the test case? That would make it easier to investigate.
These functions are from the BLAS library. Please check if the link command line includes BLAS.
Yes this sounds good, and the remark about for what parameters it would be interesting is relevant. Remark: grouping by inverted list is implemented for the PQ "fast-scan" but not...
Unfortunately we don't have this combination of GPU and CPU, so we can't investigate.
Please ask for access, we will share.