How do you bulk-load vectors efficiently into pgvector?

-- Method 1: Multi-row INSERT (batch inserts)
INSERT INTO documents (content, embedding) VALUES
    ('Document 1', '[0.1,0.2,0.3]'),
    ('Document 2', '[0.4,0.5,0.6]'),
    ('Document 3', '[0.7,0.8,0.9]');
-- Send batches of 100-1000 rows per INSERT statement

-- Method 2: COPY command (fastest for very large datasets)
COPY documents (content, embedding)
FROM '/path/to/data.csv'
WITH (FORMAT TEXT, DELIMITER '\t');
-- CSV format: text<TAB>[v1,v2,v3,...]

-- Method 3: psycopg2 executemany or copy_from (Python)
# cur.executemany(
#     "INSERT INTO documents (content, embedding) VALUES (%s, %s)",
#     [(text1, emb1), (text2, emb2), ...]
# )

-- Method 4: Disable index during load, rebuild after
-- (Much faster for large bulk loads)

-- Step 1: Create table WITHOUT index
CREATE TABLE documents (
    id BIGSERIAL PRIMARY KEY,
    embedding VECTOR(1536)
);

-- Step 2: Bulk load all data
COPY documents FROM '/data/embeddings.csv';

-- Step 3: Increase memory and workers for faster index build
SET maintenance_work_mem = '4GB';
SET max_parallel_maintenance_workers = 8;

-- Step 4: Build the index AFTER loading all data
CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops)
WITH (m=16, ef_construction=64);