HF RL Explorer

Support halfvec vector type in PgVector for high‑dimensional embedding indexes

Support halfvec vector type in PgVector for high‑dimensional embedding indexes: a task in LegoFlow-SWE (Harbor dataset). pgvector 0.7.0 introduced a halfvec type that uses 2 bytes per dimension instead of 4. This allows indexes on embeddings up to 4 000 dimensions, compared to the current 2…

The task

pgvector 0.7.0 introduced a `halfvec` type that uses 2 bytes per dimension instead of 4. This allows indexes on embeddings up to 4 000 dimensions, compared to the current 2 000‑dimension limit of the standard `vector` type. Many large embedding models (e.g. OpenAI `text‑embedding‑3‑large` with 3072 dimensions)…

Part of Lego-X/LegoFlow-SWE.