Merge branch 'main' into rgao_add_mean_pool_eqv2

src/fairchem/core/common/flags.py

@@ -8,6 +8,7 @@
 from __future__ import annotations
 
 import argparse
+import os
 from pathlib import Path
 
 
@@ -48,7 +49,7 @@ def add_core_args(self) -> None:
         )
         self.parser.add_argument(
             "--run-dir",
-            default="./",
+            default=os.path.abspath("./"),
             type=str,
             help="Directory to store checkpoint/log/result directory",
         )

src/fairchem/core/models/equiformer_v2/equiformer_v2.py

@@ -397,6 +397,9 @@ def forward(self, data: Batch) -> dict[str, torch.Tensor]:
         self.dtype = data.pos.dtype
         self.device = data.pos.device
         atomic_numbers = data.atomic_numbers.long()
+        assert (
+            atomic_numbers.max().item() < self.max_num_elements
+        ), "Atomic number exceeds that given in model config"
         graph = self.generate_graph(
             data,
             enforce_max_neighbors_strictly=self.enforce_max_neighbors_strictly,

src/fairchem/core/models/escn/escn.py

@@ -235,6 +235,9 @@ def forward(self, data):
 
         start_time = time.time()
         atomic_numbers = data.atomic_numbers.long()
+        assert (
+            atomic_numbers.max().item() < self.max_num_elements
+        ), "Atomic number exceeds that given in model config"
         num_atoms = len(atomic_numbers)
         graph = self.generate_graph(data)
 

tests/core/e2e/test_s2ef.py

@@ -170,6 +170,26 @@ def test_use_pbc_single(self, configs, tutorial_val_src, torch_deterministic):
                 input_yaml=configs["equiformer_v2"],
             )
 
+    def test_max_num_atoms(self, configs, tutorial_val_src, torch_deterministic):
+        with tempfile.TemporaryDirectory() as tempdirname:
+            tempdir = Path(tempdirname)
+            extra_args = {"seed": 0}
+            with pytest.raises(AssertionError):
+                _ = _run_main(
+                    rundir=str(tempdir),
+                    update_dict_with={
+                        "optim": {"max_epochs": 1},
+                        "model": {"backbone": {"max_num_elements": 2}},
+                        "dataset": oc20_lmdb_train_and_val_from_paths(
+                            train_src=str(tutorial_val_src),
+                            val_src=str(tutorial_val_src),
+                            test_src=str(tutorial_val_src),
+                        ),
+                    },
+                    update_run_args_with=extra_args,
+                    input_yaml=configs["equiformer_v2_hydra"],
+                )
+
     @pytest.mark.parametrize(
         ("world_size", "ddp"),
         [