Multivariate TS as Univariate with Features?

[NudnikShpilkis]

I have 1M+ multivariate items that I need to forecast. Each item has 3 features. The data looks like this:

Month	Item	Feat A	Feat B	Feat C
Jan	X	10	0.5	0
Feb	X	12	0.75	0
Mar	X	15	0.25	0.01
Apr	X	13	1	0

I'd like to try modeling the features in a univariate layout because while I don't think there are dependencies between the features within an item, there are general learning we can make about features A, B and C.

Month	Item	Feature	Y
Jan	X	A	10
Feb	X	A	12
Mar	X	A	13
Apr	X	A	15
Jan	X	B	0.5
Feb	X	B	0.75
Mar	X	B	0.25
Apr	X	B	1
Jan	X	C	0
Feb	X	C	0
Mar	X	C	0.01
Apr	X	C	0

How can I load and forecast my data in this layout in gluonts? Item ID is required to be a string and not multiple columns, so I'd be forced to do item_feature which obscures the point of including feature in the first place.