A question about reference result in readme

[Yi-sir]

For input weights [ 20, 107, 104, 64, 19, 197, 187, 157, 172, 86, 16, 27] supported in readme as example,
the reference result is [ 7, 10, 6, 8, 6, 11, 8, 9, 2, 4, 5, 1, 5, 0, 3, 1]

Why there is only one No.7 Expert(with weight 157) in phy2log?

No.7 expert' s weight is 157 and it is the 4-th biggest one, so I think it should be replicated?

Thank you!

[Yi-sir]

I think the result may be like [[5, 3], [5, 11], [6, 9], [6, 4], [8, 2], [8, 10], [7, 1], [7, 0]], is it right?

[ztxdcyy]

You need to consider the load balancing situation between nodes. In fact, our choice of replicating experts is based on the load situation within the nodes, rather than the load situation across nodes. It is recommended that you go through the process of the program execution again following the example, and pay special attention to the changes in the data before and after the execution of the balanced_packing function.