Skip to main content

ENTITY_EMBED_CACHE_MAX_ENTRIES

Constant ENTITY_EMBED_CACHE_MAX_ENTRIES 

Source
pub const ENTITY_EMBED_CACHE_MAX_ENTRIES: usize = 10_000;
Expand description

DEFAULT entry ceiling for the process-wide entity-embedding cache.

The cache used to be an unbounded HashMap: a long ingest over a corpus with many distinct entity names grew it for the whole invocation with no eviction, so its memory was bounded only by the corpus. At 1024 dims one vector costs ~4 KiB, so 10 000 entries is ~40 MiB — the point where the cache stops paying for itself.

Read it through entity_embed_cache_max_entries, never directly.