Sampler initiation stuck

One thing that jumps immediately is that you have way more parameters than observations, I suspect your model is hard to sample regardless of how fast the logp evaluation is