08:19 tomeu: uff, finally managed to decode my first weights
08:43 pH5: awesome! how does it work?
08:44 tomeu: not that well yet, but two of the values are definitely good
08:44 tomeu: I'm implementing the decoding in python, as an intermediate step towards writing the decoder in mesa
08:45 tomeu: I will put the script in mesa, I think we will be using it for quite some time still
08:49 pH5: nice, and it will only get better from here :)
08:50 tomeu: it's definitely some progress :)
08:50 pH5: should that script be for all encodings (i.e. vipnano-qi and si+ as well?)
08:51 tomeu: no, only for huffman, but I already have one for the previous method
08:51 tomeu: a bit messy, though
08:52 pH5: either way, I think having something that can decode the mesa-compressed-*.bin dumps in a central place is a good idea.
08:52 tomeu: yeah, definitely
08:53 pH5: s/central/well known/ anyway