A differentiable function from binary integer to one-hot representations

I would like to define a differentiable function $f:\{0,1\}^{\log v}\to\{0,1\}^{v}$ that converts binary number representations of $\log v$ bits into one-hot vectors. This can be accomplished by using fuzzy logic operators to convert $$f(x)_i=\bm1[i=x]$$ into $$f(x)_i = \prod_{j=1}^{\log v}(i_jx_j)+((1-i_j)(1-x_j))-(i_jx_j)((1-i_j)(1-x_j))$$ using definitions of product $\top$-norms and $\top$-conorms and the fact that $$\bm1[a=b]=(a\land b)\lor(\lnot a\land\lnot b).$$

Why did I ask this? My previous post explored a generalized version of the cross-entropy minimization assumption from my recent paper. This function could be used to make that assumption about the input IDs $h$ to a language model while keeping the dimension of $\nabla\texttt{logits}(h)$ small.