any index for IP inner product ？

[ucas010]

in the wiki,
i find the index all for L2 distance , but the datasets(vectors) have been L2 norm , it's not good to use L2 distance ,
so , any index for IP distance for 100M ~1B datasets ?
thx

[mnorris11]

I don't think I fully understood the question, but you can specify the metric you want to use (including Inner Product) when creating a FAISS index.

[ucas010]

In my vectors , processing with L2 norm , then use
faiss.index_factory(768, "PCA128,IVF16384_HNSW64,Flat")
to get the index ,
but the results are the same with different query for recall
and the distances are as follow,
"dis": [
0.2299024760723114,
0.2320370078086853,
0.233591690659523,
0.23407505452632904,
0.2347213327884674,
0.23807403445243835,
0.24065659940242767,
0.24324321746826172,
0.24762040376663208,
0.25289610028266907
],

when I set the metric with
index = faiss.index_factory(768, "PCA128,IVF16384_HNSW64,Flat",faiss.METRIC_INNER_PRODUCT)
the distances
"dis": [
0.23751525580883026,
0.23581349849700928,
0.23538577556610107,
0.23496568202972412,
0.23209497332572937,
0.23179949820041656,
0.23018941283226013,
0.22953224182128906,
0.229049414396286,
0.22857823967933655
],

and the results are not similar with query ,either