Method preload

lightllm/server/multimodal_params.py:31–51

(self, request: Request)

        self.extra_params = {}

    async def preload(self, request: Request):
        try:
            if self._type == "url":
                timeout = int(os.getenv("REQUEST_TIMEOUT", "5"))
                proxy = os.getenv("REQUEST_PROXY", None)
                audio_data = await fetch_resource(self._data, request, timeout=timeout, proxy=proxy)
            elif self._type == "base64":
                audio_data = base64.b64decode(self._data)
            else:
                raise ValueError(f"cannot read audio which type is {self._type}!")

            # check if valid audio bytes
            audio_values, _ = librosa.load(BytesIO(audio_data), sr=16000)
            from lightllm.models.whisper.defaults import MIN_AUDIO_LEN

            self.audio_length = max(audio_values.shape[0], MIN_AUDIO_LEN)  # 如果音频过短，会被pad到480的长度
            self._preload_data = audio_data
            return

        except Exception as e:
            raise ValueError(f"Failed to read image type={self._type}, data[:100]={self._data[:100]}: {e}!")

    def read(self):
        assert self._preload_data is not None