OpenAI's text-embedding-3-large model not compatible with pgvector 5.1

Problem

OpenAI says its new "text-embedding-3-large" embedding model with 3072 dimensions is the "new best performing model" with significantly higher score than "ada v2". See: https://openai.com/blog/new-embedding-models-and-api-updates I use only HNSW indexing with cosine distance. I gave it a try by modifying the src/hnsw.h as follows: -#define HNSW_MAX_DIM 2000 +#define HNSW_MAX_DIM 8192 But when insert records, it gives following error: Database error: failed to add index item to "xyz_cosine_idx" The "failed to add index item to" are only referenced in following files: ./src/ivfbuild.c ./src/ivfinsert.c ./src/hnswinsert.c ./src/hnswvacuum.c ./src/hnswbuild.c They point to BlockNumber and OffsetNumber based functions which I'm not familiar with. If someone could develop a patch against pgvector 5.1, I can give it a try and update here. Best regards Sagara