nnfp/benchmark/qps_test.py

import towhee
from towhee.dc2 import pipe, ops
from towhee import triton_client

import onnxruntime
import numpy
import torch
from statistics import mean

import time
import argparse

import os
import re
import warnings
import logging

warnings.filterwarnings("ignore", category=UserWarning)

parser = argparse.ArgumentParser()
parser.add_argument('--pipe', action='store_true')
parser.add_argument('--triton', action='store_true')
parser.add_argument('--onnx', action='store_true')
parser.add_argument('--atol', type=float, default=1e-3)
parser.add_argument('--num', type=int, default=100)
parser.add_argument('--device', type=str, default='cpu')
args = parser.parse_args()

p = (
    pipe.input('path')
        .map('path', 'frame', ops.audio_decode.ffmpeg())
        .map('frame', 'vecs', ops.audio_embedding.nnfp(device=args.device))
        .output('vecs')
)

data = '1sec.wav'
out1 = p(data).get()[0]
print('Pipe: OK', out1.shape)

if args.num and args.pipe:
    qps = []
    for _ in range(10):
        start = time.time()
        # for _ in range(args.num):
        #     p(data)
        p.batch([data] * args.num)
        end = time.time()
        q = args.num / (end - start)
        qps.append(q)
    print('Pipe qps:', mean(qps))

if args.triton:
    client = triton_client.Client(url='localhost:8000')
    out2 = client(data)[0][0]
    print('Triton: OK')

    if numpy.allclose(out1, out2, atol=args.atol):
        print('Check accuracy: OK')
    else:
        max_diff = numpy.abs(out1 - out2).max()
        min_diff = numpy.abs(out1 - out2).min()
        mean_diff =  numpy.abs(out1 - out2).mean()
        print(f'Check accuracy: atol is larger than {args.atol}.')
        print(f'Maximum absolute difference is {max_diff}.')
        print(f'Minimum absolute difference is {min_diff}.')
        print(f'Mean difference is {mean_diff}.')
    
    if args.num:
        qps = []
        for _ in range(10):
            start = time.time()
            client.batch([data] * args.num)
            end = time.time()
            q = args.num / (end - start)
            qps.append(q)
        print('Triton qps:', mean(qps))

if args.onnx:
    op = ops.audio_embedding.nnfp(device=args.device).get_op()
    decoder = ops.audio_decode.ffmpeg()
    # if not os.path.exists('test.onnx'):
    op.save_model('onnx', 'test.onnx')
    sess = onnxruntime.InferenceSession('test.onnx',
                                        providers=['CUDAExecutionProvider'])
    inputs = [x for x in decoder(data)]
    inputs = op.preprocess(inputs)
    out3 = sess.run(None, input_feed={'input': inputs.cpu().detach().numpy()})
    print('Onnx: OK', out3[0].shape)
    if numpy.allclose(out1, out3, atol=args.atol):
        print('Check accuracy: OK')
    else:
        max_diff = numpy.abs(out1 - out3).max()
        min_diff = numpy.abs(out1 - out3).min()
        mean_diff =  numpy.abs(out1 - out3).mean()
        print(f'Check accuracy: atol is larger than {args.atol}.')
        print(f'Maximum absolute difference is {max_diff}.')
        print(f'Minimum absolute difference is {min_diff}.')
        print(f'Mean difference is {mean_diff}.')

    if args.num:
        qps = []
        for _ in range(10):
            start = time.time()
            for _ in range(args.num):
                inputs = [x for x in decoder(data)]
                inputs = op.preprocess(inputs)
                sess.run(None, input_feed={'input': inputs.cpu().detach().numpy()})
            end = time.time()
            q = args.num / (end - start)
            qps.append(q)
        print('Onnx qps:', mean(qps))