stella_en_1.5B_v5, smaller dimensionality: 1024 dimensionality

This model Supports multiple different dimensions: 512, 768, 1024, 2048, 4096, 6144 and 8192.

Generally speaking, 1024d is good enough. The MTEB score of 1024d is only 0.001 lower than 8192d.

The reason why I request a lower dimensionality is that the vectorized data with 1024d consumes a significantly lower amount of space when in the VectorDB vs. 8192d.

Yet this model supports large context window, it is cheaper than OpenAI’s text-embeddings3 yet better quality even in 1024d, and with lower dimensionality could consume much less space in the vector DB.