Talk:FAISS
Requested move 22 January 2025
[edit]
 | It has been proposed in this section that FAISS be renamed and moved to Faiss. A bot will list this discussion on the requested moves current discussions subpage within an hour of this tag being placed. The discussion may be closed 7 days after being opened, if consensus has been reached (see the closing instructions). Please base arguments on article title policy, and keep discussion succinct and civil. Please use {{subst:requested move}}. Do not use {{requested move/dated}} directly. |
FAISS → Faiss – The team at Meta has aligned that just capitalizing the first letter is the correct name. That is the way it is named in the original paper. I have updated the references in the page to be "Faiss" already but I cannot change the title to "Faiss". I believe this should replace the existing redirect.
The strongest argument I can make is: Even though other sources are using it incorrectly as FAISS, the primary updated research paper https://arxiv.org/pdf/2401.08281 and the actively maintained Github repository use Faiss. At what point do we follow inconsistent sources versus the original source of truth? The original paper uses Faiss everywhere, and the Github uses Faiss everywhere, and the authors have said that it is intended to be Faiss.
Inconsistent sources:
- source 17 "FAISS vector codecs" is not the right title, it is simply "Vector codecs" (from the official Github, which uses "Faiss").
- 27 and 28 ANN bench repositories are inconsistent, some FAISS and some faiss
- source 29 "Use a FAISS vector database with Haystack" uses Faiss and FAISS inconsistently.
- source 30 "FAISS integration with Langchain" when following the URL actually uses "Faiss" in the title, but uses a mixture of FAISS, Faiss, faiss throughout the page.
Sources that use it correctly as Faiss:
- 1 through 5, 15 (papers or sources by original authors of Faiss)
- 26: "Results of the Big ANN: NeurIPS'23 competition"
Sources using it incorrectly as FAISS:
- 11: "Quicker ADC : Unlocking the Hidden Potential of Product Quantization With SIMD"
- 22: "Amazon OpenSearch Service now supports efficient vector query filters for FAISS" (but this is a web page that can be updated)
- 23: "Milvus Knowhere" (but we can work with them to update it, because we meet with them often)
Mnorris1921 (talk) 23:26, 21 January 2025 (UTC) This is a contested technical request (permalink). Mnorris1921 (talk) 20:36, 22 January 2025 (UTC)