with tf.io.gfile.GFile(filename, 'rb+') as f:
            cctx = zstandard.ZstdDecompressor()
            reader_stream = io.BufferedReader(cctx.stream_reader(f))
            reader = jsonlines.Reader(reader_stream, loads=json_parser)
            for item in reader:
                result = dict()
                if isinstance(item, str):
                    result['text'] = item
                else:
                    text = item['text']
                    if isinstance(text, list):
                        text = self.para_joiner.join(text)
                        result['text'] = text
                yield result