SVM model is not reproducible

[MaxenceGollier]

xk.txt

with this file,

using DelimitedFiles, RegularizedProblems, MLDatasets, LinearAlgebra, NLPModels
Random.seed!(1234)
model, nls_model, _ = RegularizedProblems.svm_train_model()
xk = vec(readdlm("xk.txt"))

BLAS.set_num_threads(6)
println("value with 6 threads : $(obj(model, xk))")

BLAS.set_num_threads(1)
println("value with 1 thread : $(obj(model, xk))")

returns

value with 6 threads : 2917.462526583066
value with 1 thread :  2917.4625265830664

error like these propagate and eventually make unreproducible results.

[molhab]

Same for me

julia> xk = vec(readdlm("paper/examples/xk.txt"));

julia> BLAS.set_num_threads(6)

julia> println("value with 6 threads : $(obj(model, xk))")
value with 6 threads : 2917.462526583066

julia> BLAS.set_num_threads(1)

julia> println("value with 1 thread : $(obj(model, xk))")
value with 1 thread : 2917.4625265830664