VoVAllen

7 ms·

31.
by VoVAllen 2y ago
The IVF indexing can be considered into two phases, computing the centroids (KMeans), and assigning each point to the centroids as the inverted lists. The most time-consuming part is at the KMeans stage, and can be greatly accelerated with
32.
Lama3-V project from a Stanford team plagiarized a lot from MiniCPM-Llama3-v2.5(twitter.com)
7 points by VoVAllen 2y ago|1 comments
33.
by VoVAllen 2y ago
Did you create index on the tsvector?
34.
by VoVAllen 2y ago
Amazon has open sourced their policy engine https://github.com/cedar-policy/cedar
35.
by VoVAllen 2y ago
Is it possible to be integrated with DSPy?
36.
by VoVAllen 2y ago
immich did this perfectly. see https://www.reddit.com/r/immich/comments/1bdi3dz/immich_is_a...
37.
by VoVAllen 2y ago
Why now?
38.
by VoVAllen 3y ago
Better to have an AI agent which can add the company by one-click with name (even better if it can update with the latest news)
39.
by VoVAllen 3y ago
Sift 1M is too small to make meaningful comparisons. Storing 96 floats * 1M only takes up 800Mb of memory.
40.
by VoVAllen 3y ago
It's kind of a tradeoff. Performance is just one factor when choosing the vector database. In pgvecto.rs https://github.com/tensorchord/pgvecto.rs , we store the index separately from PostgreSQL's internal sto
41.
by VoVAllen 3y ago
Hi, we've solved the problem you mentioned! Please take a look on our open source postgres vector extension https://github.com/tensorchord/pgvecto.rs . Our index building process is significantly faster than pgvect
42.
by VoVAllen 3y ago
Why it's a yellow flag?
43.
by VoVAllen 3y ago
Why did you choose SpeedANN instead of other new indexes such as DiskANN? And you changed the color of epsilla in every benchmark figure, which is quite confusing
44.
by VoVAllen 3y ago
- We used a different index method called HNSW, which is more widely used in vector search area and also faster than ivfflat used by pgvector. - There are some drawbacks with HNSW. It's designed for memory usage but not so suitable as
45.
by VoVAllen 3y ago
Your homepage looks quite similar to modal.com
46.
by VoVAllen 4y ago
It's all about features and data scale. Recommendation system itself is actually a large table, DL method already proved effectiveness there. Let's say if you have text in your tabular data. Tree model(with traditional method such