Ok Thanks.
Is there a way also to save the result/status of
optimizer = numpyro.optim.Adam(exponential_decay(5e-3,1000,0.1, end_value=1e-7))
svi = SVI(model_spl, guide,optimizer,loss=Trace_ELBO())
svi_result = svi.run(jax.random.PRNGKey(0), 1000, cl_obs)
to perform a new optimisation from the last one?
As well as save the status for every new loss decrease?