Reduce blackjax sampling memory usage

conda-envs/environment-jax.yml

@@ -11,9 +11,9 @@ dependencies:
 - cloudpickle
 - h5py>=2.7
 # Jaxlib version must not be greater than jax version!
-- blackjax==1.2.0  # Blackjax>=1.2.1 is incompatible with latest available version of jaxlib in conda-forge
-- jaxlib==0.4.23  # Latest available version in conda-forge, update when new version is available
-- jax==0.4.23
+- blackjax>=1.2.2
+- jax>=0.4.28
+- jaxlib>=0.4.28
 - libblas=*=*mkl
 - mkl-service
 - numpy>=1.15.0
@@ -25,9 +25,8 @@ dependencies:
 - networkx
 - rich>=13.7.1
 - threadpoolctl>=3.1.0
-# JAX is only compatible with Scipy 1.13.0 from >=0.4.26, but the respective version of
-# JAXlib is still not on conda: https://github.com/conda-forge/jaxlib-feedstock/pull/243
-- scipy>=1.4.1,<1.13.0
+# JAX is only compatible with Scipy 1.13.0 from >=0.4.26
+- scipy>=1.13.0
 - typing-extensions>=3.7.4
 # Extra dependencies for testing
 - ipython>=7.16

pymc/sampling/jax.py

@@ -243,6 +243,8 @@ def _blackjax_inference_loop(
 ):
     import blackjax
 
+    from blackjax.adaptation.base import get_filter_adapt_info_fn
+
     algorithm_name = adaptation_kwargs.pop("algorithm", "nuts")
     if algorithm_name == "nuts":
         algorithm = blackjax.nuts
@@ -255,6 +257,7 @@ def _blackjax_inference_loop(
         algorithm=algorithm,
         logdensity_fn=logprob_fn,
         target_acceptance_rate=target_accept,
+        adaptation_info_fn=get_filter_adapt_info_fn(),
         **adaptation_kwargs,
     )
     (last_state, tuned_params), _ = adapt.run(seed, init_position, num_steps=tune)

tests/sampling/test_mcmc_external.py

@@ -51,7 +51,7 @@ def test_external_nuts_sampler(recwarn, nuts_sampler):
     warns = {
         (warn.category, warn.message.args[0])
         for warn in recwarn
-        if warn.category not in (FutureWarning, DeprecationWarning)
+        if warn.category not in (FutureWarning, DeprecationWarning, RuntimeWarning)
     }
     expected = set()
     if nuts_sampler == "nutpie":