from towhee import triton_client
import sys
import time

num = int(sys.argv[-1])
data = 'Hello, world.'
client = triton_client.Client('localhost:8000')

# warm up
client.batch([data])
print('client: ok')

time.sleep(5)

print('test...')
start = time.time()
client.batch([data] * num, batch_size=8)
end = time.time()
print(f'duration: {end - start}')
print(f'qps: {num / (end - start)}')