Drop the support for PyTorch<2.0 (#3272)

CONTRIBUTING.md

@@ -44,7 +44,7 @@ To run a single test from the command line
 ```sh
 pytest -vs {path_to_test}::{test_name}
 # or in cuda mode
-CUDA_TEST=1 PYRO_TENSOR_TYPE=torch.cuda.DoubleTensor pytest -vs {path_to_test}::{test_name}
+CUDA_TEST=1 PYRO_DTYPE=float64 PYRO_DEVICE=cuda pytest -vs {path_to_test}::{test_name}
 ```
 
 To ensure documentation builds correctly, run

Makefile

@@ -69,11 +69,11 @@ test-all: lint FORCE
 	  | xargs pytest -vx --nbval-lax
 
 test-cuda: lint FORCE
-	CUDA_TEST=1 PYRO_TENSOR_TYPE=torch.cuda.DoubleTensor pytest -vx --stage unit
+	CUDA_TEST=1 PYRO_DTYPE=float64 PYRO_DEVICE=cuda pytest -vx --stage unit
 	CUDA_TEST=1 pytest -vx tests/test_examples.py::test_cuda
 
 test-cuda-lax: lint FORCE
-	CUDA_TEST=1 PYRO_TENSOR_TYPE=torch.cuda.DoubleTensor pytest -vx --stage unit --lax
+	CUDA_TEST=1 PYRO_DTYPE=float64 PYRO_DEVICE=cuda pytest -vx --stage unit --lax
 	CUDA_TEST=1 pytest -vx tests/test_examples.py::test_cuda
 
 test-jit: FORCE

docs/source/conf.py

@@ -223,6 +223,6 @@ def setup(app):
 if "READTHEDOCS" in os.environ:
     os.system("pip install numpy")
     os.system(
-        "pip install torch==1.11.0+cpu torchvision==0.12.0+cpu "
+        "pip install torch==2.0+cpu torchvision==0.15.0+cpu "
         "-f https://download.pytorch.org/whl/torch_stable.html"
     )

examples/baseball.py

@@ -418,6 +418,6 @@ def main(args):
     torch.multiprocessing.set_sharing_strategy("file_system")
 
     if args.cuda:
-        torch.set_default_tensor_type(torch.cuda.FloatTensor)
+        torch.set_default_device("cuda")
 
     main(args)

examples/contrib/cevae/synthetic.py

@@ -45,7 +45,7 @@ def generate_data(args):
 
 def main(args):
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
+        torch.set_default_device("cuda")
 
     # Generate synthetic data.
     pyro.set_rng_seed(args.seed)

examples/contrib/epidemiology/regional.py

@@ -205,12 +205,9 @@ def main(args):
     if args.warmup_steps is None:
         args.warmup_steps = args.num_samples
     if args.double:
-        if args.cuda:
-            torch.set_default_tensor_type(torch.cuda.DoubleTensor)
-        else:
-            torch.set_default_dtype(torch.float64)
-    elif args.cuda:
-        torch.set_default_tensor_type(torch.cuda.FloatTensor)
+        torch.set_default_dtype(torch.float64)
+    if args.cuda:
+        torch.set_default_device("cuda")
 
     main(args)
 

examples/contrib/epidemiology/sir.py

@@ -391,12 +391,9 @@ def main(args):
     if args.warmup_steps is None:
         args.warmup_steps = args.num_samples
     if args.double:
-        if args.cuda:
-            torch.set_default_tensor_type(torch.cuda.DoubleTensor)
-        else:
-            torch.set_default_dtype(torch.float64)
-    elif args.cuda:
-        torch.set_default_tensor_type(torch.cuda.FloatTensor)
+        torch.set_default_dtype(torch.float64)
+    if args.cuda:
+        torch.set_default_device("cuda")
 
     main(args)
 

examples/contrib/funsor/hmm.py

@@ -670,7 +670,7 @@ def model_7(sequences, lengths, args, batch_size=None, include_prior=True):
 
 def main(args):
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
+        torch.set_default_device("cuda")
 
     logging.info("Loading data")
     data = poly.load_data(poly.JSB_CHORALES)

examples/contrib/mue/FactorMuE.py

@@ -427,9 +427,8 @@ def main(args):
     )
     args = parser.parse_args()
 
+    torch.set_default_dtype(torch.float64)
     if args.cuda:
-        torch.set_default_tensor_type(torch.cuda.DoubleTensor)
-    else:
-        torch.set_default_dtype(torch.float64)
+        torch.set_default_device("cuda")
 
     main(args)

examples/contrib/mue/ProfileHMM.py

@@ -316,9 +316,8 @@ def main(args):
     )
     args = parser.parse_args()
 
+    torch.set_default_dtype(torch.float64)
     if args.cuda:
-        torch.set_default_tensor_type(torch.cuda.DoubleTensor)
-    else:
-        torch.set_default_dtype(torch.float64)
+        torch.set_default_device("cuda")
 
     main(args)

examples/einsum.py

@@ -174,9 +174,7 @@ def time_fn(fn, equation, *operands, **kwargs):
 
 def main(args):
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
-    else:
-        torch.set_default_tensor_type("torch.FloatTensor")
+        torch.set_default_device("cuda")
 
     if args.method == "all":
         for method in ["prob", "logprob", "gradient", "marginal", "map", "sample"]:

examples/hmm.py

@@ -620,7 +620,7 @@ def model_7(sequences, lengths, args, batch_size=None, include_prior=True):
 
 def main(args):
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
+        torch.set_default_device("cuda")
 
     logging.info("Loading data")
     data = poly.load_data(poly.JSB_CHORALES)

examples/sir_hmc.py

@@ -663,12 +663,9 @@ def main(args):
     args = parser.parse_args()
 
     if args.double:
-        if args.cuda:
-            torch.set_default_tensor_type(torch.cuda.DoubleTensor)
-        else:
-            torch.set_default_tensor_type(torch.DoubleTensor)
-    elif args.cuda:
-        torch.set_default_tensor_type(torch.cuda.FloatTensor)
+        torch.set_default_dtype(torch.float64)
+    if args.cuda:
+        torch.set_default_device("cuda")
 
     main(args)
 

examples/sparse_gamma_def.py

@@ -31,7 +31,7 @@
 from pyro.infer import SVI, TraceMeanField_ELBO
 from pyro.infer.autoguide import AutoDiagonalNormal, init_to_feasible
 
-torch.set_default_tensor_type("torch.FloatTensor")
+torch.set_default_dtype(torch.float32)
 pyro.util.set_rng_seed(0)
 
 

examples/sparse_regression.py

@@ -41,7 +41,7 @@
 """
 
 
-torch.set_default_tensor_type("torch.FloatTensor")
+torch.set_default_dtype(torch.float32)
 
 
 def dot(X, Z):

examples/svi_horovod.py

@@ -78,7 +78,7 @@ def main(args):
         if args.cuda:
             torch.cuda.set_device(hvd.local_rank())
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
+        torch.set_default_device("cuda")
     device = torch.tensor(0).device
 
     if args.horovod:

examples/svi_lightning.py

@@ -15,7 +15,7 @@
 
 import argparse
 
-import pytorch_lightning as pl
+import lightning.pytorch as pl
 import torch
 
 import pyro

profiler/gaussianhmm.py

@@ -21,7 +21,7 @@ def random_mvn(batch_shape, dim, requires_grad=False):
 
 def main(args):
     if args.cuda:
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
+        torch.set_default_device("cuda")
 
     hidden_dim = args.hidden_dim
     obs_dim = args.obs_dim

pyro/contrib/gp/parameterized.py

@@ -82,7 +82,7 @@ class Parameterized(PyroModule):
         >>> assert "b_scale_unconstrained" in dict(linear.named_parameters())
 
     Note that by default, data of a parameter is a float :class:`torch.Tensor`
-    (unless we use :func:`torch.set_default_tensor_type` to change default
+    (unless we use :func:`torch.set_default_dtype` to change default
     tensor type). To cast these parameters to a correct data type or GPU device,
     we can call methods such as :meth:`~torch.nn.Module.double` or
     :meth:`~torch.nn.Module.cuda`. See :class:`torch.nn.Module` for more

pyro/infer/mcmc/api.py

@@ -107,13 +107,15 @@ def __init__(
         self.rng_seed = (torch.initial_seed() + chain_id) % MAX_SEED
         self.log_queue = log_queue
         self.result_queue = result_queue
-        self.default_tensor_type = torch.Tensor().type()
+        self.default_dtype = torch.Tensor().dtype
+        self.default_device = torch.Tensor().device
         self.hook = hook
         self.event = event
 
     def run(self, *args, **kwargs):
         pyro.set_rng_seed(self.rng_seed)
-        torch.set_default_tensor_type(self.default_tensor_type)
+        torch.set_default_dtype(self.default_dtype)
+        torch.set_default_device(self.default_device)
         kwargs = kwargs
         logger = logging.getLogger("pyro.infer.mcmc")
         logger_id = "CHAIN:{}".format(self.chain_id)

pyro/ops/provenance.py

@@ -93,11 +93,6 @@ def _track_provenance_set(x, provenance: frozenset):
 @track_provenance.register(tuple)
 @track_provenance.register(dict)
 def _track_provenance_pytree(x, provenance: frozenset):
-    # avoid max-recursion depth error for torch<=2.0
-    flat_args, _ = tree_flatten(x)
-    if not flat_args or flat_args[0] is x:
-        return x
-
     return tree_map(partial(track_provenance, provenance=provenance), x)
 
 
@@ -143,11 +138,6 @@ def _extract_provenance_set(x):
 @extract_provenance.register(tuple)
 @extract_provenance.register(dict)
 def _extract_provenance_pytree(x):
-    # avoid max-recursion depth error for torch<=2.0
-    flat_args, _ = tree_flatten(x)
-    if not flat_args or flat_args[0] is x:
-        return x, frozenset()
-
     flat_args, spec = tree_flatten(x)
     xs = []
     provenance = frozenset()

pyro/optim/pytorch_optimizers.py

@@ -34,11 +34,7 @@
     del _PyroOptim
 
 # Load all schedulers from PyTorch
-# breaking change in torch >= 1.14: LRScheduler is new base class
-if hasattr(torch.optim.lr_scheduler, "LRScheduler"):
-    _torch_scheduler_base = torch.optim.lr_scheduler.LRScheduler  # type: ignore
-else:  # for torch < 1.13, _LRScheduler is base class
-    _torch_scheduler_base = torch.optim.lr_scheduler._LRScheduler  # type: ignore
+_torch_scheduler_base = torch.optim.lr_scheduler.LRScheduler  # type: ignore
 
 for _name, _Optim in torch.optim.lr_scheduler.__dict__.items():
     if not isinstance(_Optim, type):

setup.py

@@ -68,10 +68,10 @@
     "jupyter>=1.0.0",
     "graphviz>=0.8",
     "matplotlib>=1.3",
-    "torchvision>=0.12.0",
+    "torchvision>=0.15.0",
     "visdom>=0.1.4,<0.2.2",  # FIXME visdom.utils is unavailable >=0.2.2
     "pandas",
-    "pillow==8.2.0",  # https://github.com/pytorch/pytorch/issues/61125
+    "pillow>=8.3.1",  # https://github.com/pytorch/pytorch/issues/61125
     "scikit-learn",
     "seaborn>=0.11.0",
     "wget",
@@ -102,7 +102,7 @@
         "numpy>=1.7",
         "opt_einsum>=2.3.2",
         "pyro-api>=0.1.1",
-        "torch>=1.11.0",
+        "torch>=2.0",
         "tqdm>=4.36",
     ],
     extras_require={
@@ -135,7 +135,7 @@
             "yapf",
         ],
         "horovod": ["horovod[pytorch]>=0.19"],
-        "lightning": ["pytorch_lightning"],
+        "lightning": ["lightning"],
         "funsor": [
             # This must be a released version when Pyro is released.
             # "funsor[torch] @ git+git://github.com/pyro-ppl/funsor.git@7bb52d0eae3046d08a20d1b288544e1a21b4f461",

tests/common.py

@@ -69,11 +69,11 @@ def wrapper(*args, **kwargs):
 )
 
 try:
-    import pytorch_lightning
+    import lightning
 except ImportError:
-    pytorch_lightning = None
+    lightning = None
 requires_lightning = pytest.mark.skipif(
-    pytorch_lightning is None, reason="pytorch lightning is not available"
+    lightning is None, reason="pytorch lightning is not available"
 )
 
 try:
@@ -93,23 +93,6 @@ def get_gpu_type(t):
     return getattr(torch.cuda, t.__name__)
 
 
-@contextlib.contextmanager
-def tensors_default_to(host):
-    """
-    Context manager to temporarily use Cpu or Cuda tensors in PyTorch.
-
-    :param str host: Either "cuda" or "cpu".
-    """
-    assert host in ("cpu", "cuda"), host
-    old_module, name = torch.Tensor().type().rsplit(".", 1)
-    new_module = "torch.cuda" if host == "cuda" else "torch"
-    torch.set_default_tensor_type("{}.{}".format(new_module, name))
-    try:
-        yield
-    finally:
-        torch.set_default_tensor_type("{}.{}".format(old_module, name))
-
-
 @contextlib.contextmanager
 def default_dtype(dtype):
     """

tests/conftest.py

@@ -9,7 +9,9 @@
 
 import pyro
 
-torch.set_default_tensor_type(os.environ.get("PYRO_TENSOR_TYPE", "torch.DoubleTensor"))
+DTYPE = getattr(torch, os.environ.get("PYRO_DTYPE", "float64"))
+torch.set_default_dtype(DTYPE)
+torch.set_default_device(os.environ.get("PYRO_DEVICE", "cpu"))
 
 
 def pytest_configure(config):

tests/contrib/timeseries/test_gp.py

@@ -40,7 +40,7 @@
 )
 @pytest.mark.parametrize("T", [11, 37])
 def test_timeseries_models(model, nu_statedim, obs_dim, T):
-    torch.set_default_tensor_type("torch.DoubleTensor")
+    torch.set_default_dtype(torch.float64)
     dt = 0.1 + torch.rand(1).item()
 
     if model == "lcmgp":

tests/contrib/timeseries/test_lgssm.py

@@ -13,7 +13,7 @@
 @pytest.mark.parametrize("obs_dim", [2, 4])
 @pytest.mark.parametrize("T", [11, 17])
 def test_generic_lgssm_forecast(model_class, state_dim, obs_dim, T):
-    torch.set_default_tensor_type("torch.DoubleTensor")
+    torch.set_default_dtype(torch.float64)
 
     if model_class == "lgssm":
         model = GenericLGSSM(