@cliffwang
Do you know any free software to merge HDR video?
No.
If you already have a half decent NLE you probably have the tools you need to merge different exposure layers (although I'm not sure that the HDR ML gives you is quite the same process as merging stills, I suspect it's already done in camera probably by making the most of the sensors dynamic range, but I don't use ML so what do I know...)
If you had say 3 identical video takes from a tripod with different exposures, then you could place each take in it's own video layer (v1, v2, v3) and either blend using different opacity levels, or composite using different modes (difference, exclusion, multiply, screen etc)
In stills mode the barrier to successful HDR is cloud, people or foilage movement, in video this would be 10 times worse, the only way I could imagine combating it to give you three seperate video expsosures would be to have the same footage pulled down at 3 different exposures, for depth of field parity and motion parity this would probably need to be 3 sensors set up around a prism, one chip at ISO 100, one at ISO 200, one at ISO 400 for example.
As such I really don't think the ML HDR mode will work like stills, it will be one file already processed, probably using selective clipping and crunching beyond the usual processed range.
Still HDR but not in the way stills guys are used to.