Creating an HNSW index causes severe data inflation

[QuanZi123]

duckdb version : 1.1.1

When create HNSW indexes first and then insert data, duckdb's data files are 100 times larger than the original files

But if insert the data first and then create the HNSW index, duckdb's data swells by about 15 times

Test data and code duckdb_index

Here's my test result:

Original file train.jsonl : 78371kb

File speed_3.db (create index first) : 7055628KB

File speed_t.db (insert data first) : 1328369KB