I would like to sort each histogram from a stacked barplot so that each column has the higher value at the bottom.

Here is an example of what i obtain that you can reproduce

library(plotly)
m <- matrix(0, ncol = 3, nrow = 20)
m <- data.frame(m)
colnames(m)= c("variable","sample","value")
m$variable <- rep(c("A","B","C","D","E"),4)
m$sample <- rep(paste("Sample",seq(1,4),sep="_"),each=5)
m$value = c(20,1,50,17,34,23,45,1,4,8,10,24,45,32,17,49,43,25,14,5)

plot_ly(m,
x = ~sample,
y = ~value,
color = ~variable,
    type = "bar")%>%
layout(barmode = 'stack')

How can i sort each column so that most abundant values are sorted (the higher at the bottom at the bottom ???)

The plotly package is a CRAN package, not Bioconductor, so you should ask your questions on the R-help listserv (r-help@r-project.org), or on the plot.ly community help.

But do note that the primary usefulness of a stacked barchart comes from having everything stacked in the same order in each bar. That's like the whole point of doing a stacked barchart, so I would be rather surprised if there was an easy way to undo the one redeeming feature of that plot.