perf: reduce per-query latency — shared model, single encode, float32 LLM

d624b44

hodfa840 commited on 16 days ago