Benchmarking Performance

import time, statistics

times = []
for qv in queries:
    t = time.perf_counter()
    index.query(vector=qv, top_k=10)
    times.append((time.perf_counter() - t) * 1000)

print("p50", statistics.median(times))
print("p95", statistics.quantiles(times, n=20)[18])
print("p99", statistics.quantiles(times, n=100)[98])

t = time.perf_counter()
total = 0
for b in batches:
    index.upsert(vectors=b)
    total += len(b)
elapsed = time.perf_counter() - t
print(f"{total/elapsed:.0f} vectors/sec")

from concurrent.futures import ThreadPoolExecutor

def run(qv):
    t = time.perf_counter()
    index.query(vector=qv, top_k=10)
    return (time.perf_counter() - t) * 1000

with ThreadPoolExecutor(max_workers=50) as ex:
    times = list(ex.map(run, queries))

from memory_profiler import profile

@profile
def upsert_all():
    for b in batches: index.upsert(vectors=b)