Pymc3 produces different results than Stan/NumPyro

I am getting no divergences and seemingly good convergence (note the \hat{r}). I just cut and pasted your code. I am running pymc v3.11.2 and theano-pymc v1.1.2.

Multiprocess sampling (4 chains in 2 jobs)
NUTS: [b, w]
Sampling 4 chains for 1_000 tune and 1_000 draw iterations (4_000 + 4_000 draws total) took 40 seconds.
The acceptance probability does not match the target. It is 0.8817572415352573, but should be close to 0.8. Try to increase the number of tuning steps.
The acceptance probability does not match the target. It is 0.8824272675568593, but should be close to 0.8. Try to increase the number of tuning steps.
The number of effective samples is smaller than 25% for some parameters.
        mean     sd  hdi_3%  hdi_97%  mcse_mean  mcse_sd  ess_bulk  ess_tail  r_hat
w[0]  10.514  8.580  -5.108   27.238      0.358    0.256     578.0     754.0    1.0
w[1] -16.138  7.215 -29.167   -1.851      0.299    0.212     585.0     735.0    1.0
w[2]   4.954  6.910  -7.890   18.009      0.289    0.205     572.0     723.0    1.0
w[3]   4.810  4.263  -3.699   12.306      0.158    0.112     733.0    1192.0    1.0
w[4]  -1.607  3.621  -8.023    5.442      0.091    0.065    1574.0    2033.0    1.0
w[5]  24.351  7.859   9.767   39.080      0.321    0.227     603.0     808.0    1.0
w[6]  19.428  7.858   4.582   33.896      0.328    0.233     577.0     729.0    1.0
w[7]   4.293  3.254  -1.798   10.347      0.126    0.089     667.0     956.0    1.0
w[8]  -3.805  3.015  -9.293    2.117      0.117    0.083     670.0     937.0    1.0
w[9]  -0.565  1.997  -4.291    3.273      0.061    0.043    1104.0    1689.0    1.0
b     -1.092  9.642 -19.659   16.161      0.385    0.273     628.0     848.0    1.0
1 Like