THe LZ4 filter code is not correct. It accepts one int argument, the level. It attempts to set the LZ4 block size to the level. This is not correct and results in a trivially small blocksize which messes up the compression. I will put up a PR with a fix.