calib set

游雁

2023-03-15 01b4ff3bde05b7c5ea071af8867a331e9ae4bf53

calib set

 funasr/export/export_model.py

@@ -21,6 +21,8 @@
        onnx: bool = True,
        quant: bool = True,
        fallback_num: int = 0,
        audio_in: str = None,
        calib_num: int = 200,
    ):
        assert check_argument_types()
        self.set_all_random_seed(0)
@@ -36,6 +38,9 @@
        self.onnx = onnx
        self.quant = quant
        self.fallback_num = fallback_num
        self.frontend = None
        self.audio_in = audio_in
        self.calib_num = calib_num
        

    def _export(
@@ -67,8 +72,14 @@
    def _torch_quantize(self, model):
        def _run_calibration_data(m):
            # using dummy inputs for a example
            if self.audio_in is not None:
                feats, feats_len = self.load_feats(self.audio_in)
                for feat, len in zip(feats, feats_len):
                    m(feat, len)
            else:
            dummy_input = model.get_dummy_inputs()
            m(*dummy_input)
            

        from torch_quant.module import ModuleFilter
        from torch_quant.quantizer import Backend, Quantizer
@@ -114,6 +125,39 @@
        random.seed(seed)
        np.random.seed(seed)
        torch.random.manual_seed(seed)

    def parse_audio_in(self, audio_in):
        
        wav_list, name_list = [], []
        if audio_in.endswith(".scp"):
            f = open(audio_in, 'r')
            lines = f.readlines()[:self.calib_num]
            for line in lines:
                name, path = line.strip().split()
                name_list.append(name)
                wav_list.append(path)
        else:
            wav_list = [audio_in,]
            name_list = ["test",]
        return wav_list, name_list
    
    def load_feats(self, audio_in: str = None):
        import torchaudio

        wav_list, name_list = self.parse_audio_in(audio_in)
        feats = []
        feats_len = []
        for line in wav_list:
            name, path = line.strip().split()
            waveform, sampling_rate = torchaudio.load(path)
            if sampling_rate != self.frontend.fs:
                waveform = torchaudio.transforms.Resample(orig_freq=sampling_rate,
                                                          new_freq=self.frontend.fs)(waveform)
            fbank, fbank_len = self.frontend(waveform, [waveform.size(1)])
            feats.append(fbank)
            feats_len.append(fbank_len)
        return feats, feats_len
    
    def export(self,
               tag_name: str = 'damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
               mode: str = 'paraformer',
@@ -190,6 +234,8 @@
    parser.add_argument('--type', type=str, default='onnx', help='["onnx", "torch"]')
    parser.add_argument('--quantize', action='store_true', help='export quantized model')
    parser.add_argument('--fallback-num', type=int, default=0, help='amp fallback number')
    parser.add_argument('--audio_in', type=str, default=None, help='["wav", "wav.scp"]')
    parser.add_argument('--calib_num', type=int, default=200, help='calib max num')
    args = parser.parse_args()

    export_model = ASRModelExportParaformer(
@@ -197,5 +243,7 @@
        onnx=args.type == 'onnx',
        quant=args.quantize,
        fallback_num=args.fallback_num,
        audio_in=args.audio_in,
        calib_num=args.calib_num,
    )
    export_model.export(args.model_name)

 funasr/export/utils/wav_load.py

New file
@@ -0,0 +1,2 @@
import os