Parallel nested graph generation attempt

[DiagRisker]

Hello,

I'm trying to speed up this simple function (with parallel scalar injection):

import jax; from jax.tree_util import Partial
@Partial(jax.jit, static_argnums=0)
def FVconnect(n: int = 7): # N being the number of Vertices 
    """ fully connect vertices between each others in a fully connected manner (densest option) """ 
    Edges = [ (i,j) for i in range(n) for j in range(i+1,n) ]
    return Edges

Output = FVconnect() # 

this will give a 2nd degree tensor (but what matters is only j in the end, i is repeated)
So I tried 2 versions focusing on j, one with jax.vmap, and another with jax.lax.scan:

def generate_edges1(n: int = 2): # N being the number of Vertices 
    Edges = jax.pmap(lambda a: jnp.arange(a+1,n), 0, None)(jnp.arange(n-1))
    return Edges

def generate_edges2(n: int = 2): # N being the number of Vertices 
    Edges = jax.lax.scan(lambda b,a: jnp.arange(a+1,n),init=0,xs=jnp.arange(n-1) )
    return Edges

In idea the array will ressemble this with jax.vmap

Output: array([ 1, 2, 3, 4, 5, 6],
[ 2, 3, 4, 5, 6],
[ 3, 4, 5, 6],
[ 4, 5, 6],
[ 5, 6],
[ 6])

However as for numpy Arrays, jax.Array type does not work with non uniform shapes, so I'm ok with flattening / concatenating this.
But jax.lax.scan doesn't seem to work with this to my surprise..

Anyone in the forum knows how to make it work?

Thanks in advance!

[DiagRisker]

Update:
see the topic #32993 (comment) where a solution is proposed along this line (for those interested)