decoding

游雁

2024-06-13 c553a8db1712c2a5deeef5bbb68bd1fdf8d61ab7

 runtime/python/utils/test_rtf_gpu.py

@@ -1,58 +1,82 @@

import time
import sys
import librosa
from funasr.utils.types import str2bool

import argparse

parser = argparse.ArgumentParser()
parser.add_argument('--model_dir', type=str, required=True)
parser.add_argument('--backend', type=str, default='onnx', help='["onnx", "torch"]')
parser.add_argument('--wav_file', type=str, default=None, help='amp fallback number')
parser.add_argument('--quantize', type=str2bool, default=False, help='quantized model')
parser.add_argument('--intra_op_num_threads', type=int, default=1, help='intra_op_num_threads for onnx')
parser.add_argument('--batch_size', type=int, default=1, help='batch_size for onnx')
parser.add_argument("--model_dir", type=str, required=True)
parser.add_argument("--backend", type=str, default="onnx", help='["onnx", "torch"]')
parser.add_argument("--wav_file", type=str, default=None, help="amp fallback number")
parser.add_argument("--quantize", type=str2bool, default=False, help="quantized model")
parser.add_argument(
    "--intra_op_num_threads", type=int, default=1, help="intra_op_num_threads for onnx"
)
parser.add_argument("--batch_size", type=int, default=1, help="batch_size for onnx")
args = parser.parse_args()


from funasr.runtime.python.libtorch.funasr_torch import Paraformer
if args.backend == "onnx":
   from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
	
model = Paraformer(args.model_dir, batch_size=args.batch_size, quantize=args.quantize, intra_op_num_threads=args.intra_op_num_threads)

wav_file_f = open(args.wav_file, 'r')
if args.backend == "onnx":
    from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer

model = Paraformer(
    args.model_dir,
    batch_size=args.batch_size,
    quantize=args.quantize,
    intra_op_num_threads=args.intra_op_num_threads,
)

wav_file_f = open(args.wav_file, "r")
wav_files = wav_file_f.readlines()

# warm-up
total = 0.0
num = 30
wav_path = wav_files[0].split("\t")[1].strip() if "\t" in wav_files[0] else wav_files[0].split(" ")[1].strip()
wav_path = (
    wav_files[0].split("\t")[1].strip()
    if "\t" in wav_files[0]
    else wav_files[0].split(" ")[1].strip()
)
for i in range(num):
   beg_time = time.time()
   result = model(wav_path)
   end_time = time.time()
   duration = end_time-beg_time
   total += duration
   print(result)
   print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
    beg_time = time.time()
    result = model(wav_path)
    end_time = time.time()
    duration = end_time - beg_time
    total += duration
    print(result)
    print(
        "num: {}, time, {}, avg: {}, rtf: {}".format(
            len(wav_path), duration, total / (i + 1), (total / (i + 1)) / 5.53
        )
    )

# infer time
wav_path = []
beg_time = time.time()
for i, wav_path_i in enumerate(wav_files):
   wav_path_i = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
   wav_path += [wav_path_i]
    wav_path_i = (
        wav_path_i.split("\t")[1].strip()
        if "\t" in wav_path_i
        else wav_path_i.split(" ")[1].strip()
    )
    wav_path += [wav_path_i]
result = model(wav_path)
end_time = time.time()
duration = (end_time-beg_time)*1000
duration = (end_time - beg_time) * 1000
print("total_time_comput_ms: {}".format(int(duration)))

duration_time = 0.0
for i, wav_path_i in enumerate(wav_files):
   wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
   waveform, _ = librosa.load(wav_path, sr=16000)
   duration_time += len(waveform)/16.0
    wav_path = (
        wav_path_i.split("\t")[1].strip()
        if "\t" in wav_path_i
        else wav_path_i.split(" ")[1].strip()
    )
    waveform, _ = librosa.load(wav_path, sr=16000)
    duration_time += len(waveform) / 16.0
print("total_time_wav_ms: {}".format(int(duration_time)))

print("total_rtf: {:.5}".format(duration/duration_time))
print("total_rtf: {:.5}".format(duration / duration_time))