Fix: finetune error with cpu device

src/mattersim/training/finetune_mattersim.py

@@ -17,11 +17,14 @@
 from mattersim.utils.logger_utils import get_logger
 
 logger = get_logger()
-torch.distributed.init_process_group(backend="nccl")
 local_rank = int(os.environ["LOCAL_RANK"])
 
 
 def main(args):
+    if args.device == "cuda":
+        torch.distributed.init_process_group(backend="nccl")
+    else:
+        torch.distributed.init_process_group(backend="gloo")
     args_dict = vars(args)
     if args.wandb and local_rank == 0:
         wandb_api_key = (
@@ -48,7 +51,8 @@ def main(args):
     np.random.seed(args.seed)
     torch.manual_seed(args.seed)
 
-    torch.cuda.set_device(local_rank)
+    if args.device == "cuda":
+        torch.cuda.set_device(local_rank)
 
     if args.train_data_path.endswith(".pkl"):
         with open(args.train_data_path, "rb") as f:
@@ -72,12 +76,12 @@ def main(args):
         forces,
         stresses,
         shuffle=True,
-        pin_memory=True,
+        pin_memory=(args.device == "cuda"),
         is_distributed=True,
         **args_dict,
     )
 
-    device = "cuda" if torch.cuda.is_available() else "cpu"
+    device = args.device
     # build energy normalization module
     if args.re_normalize:
         scale = AtomScaling(
@@ -109,7 +113,7 @@ def main(args):
             energies,
             forces,
             stresses,
-            pin_memory=True,
+            pin_memory=(args.device == "cuda"),
             is_distributed=True,
             **args_dict,
         )
@@ -125,7 +129,8 @@ def main(args):
     if args.re_normalize:
         potential.model.set_normalizer(scale)
 
-    potential.model = torch.nn.parallel.DistributedDataParallel(potential.model)
+    if args.device == "cuda":
+        potential.model = torch.nn.parallel.DistributedDataParallel(potential.model)
     torch.distributed.barrier()
 
     potential.train_model(
@@ -136,7 +141,7 @@ def main(args):
         **args_dict,
     )
 
-    if local_rank == 0 and args.save_checkpoint:
+    if local_rank == 0 and args.save_checkpoint and args.wandb:
         wandb.save(os.path.join(args.save_path, "best_model.pth"))
 
 

src/mattersim/utils/vasprun_to_xyz.py

@@ -1,37 +1,31 @@
 # -*- coding: utf-8 -*-
+import argparse
 import os
 import random
 
 from ase.io import write
 
 from mattersim.utils.atoms_utils import AtomsAdaptor
 
-vasp_files = [
-    "work/data/H/vasp/vasprun.xml",
-    "work/data/H/vasp_2/vasprun.xml",
-    "work/data/H/vasp_3/vasprun.xml",
-    "work/data/H/vasp_4/vasprun.xml",
-    "work/data/H/vasp_5/vasprun.xml",
-    "work/data/H/vasp_6/vasprun.xml",
-    "work/data/H/vasp_7/vasprun.xml",
-    "work/data/H/vasp_8/vasprun.xml",
-    "work/data/H/vasp_9/vasprun.xml",
-    "work/data/H/vasp_10/vasprun.xml",
-]
-train_ratio = 0.8
-validation_ratio = 0.1
-test_ratio = 0.1
-
-save_dir = "./xyz_files"
-os.makedirs(save_dir, exist_ok=True)
-
-
-def main():
+
+def main(args):
+    vasp_files = []
+    for root, dirs, files in os.walk(args.data_path):
+        for file in files:
+            if file.endswith(".xml"):
+                vasp_files.append(os.path.join(root, file))
+
+    train_ratio = args.train_ratio
+    validation_ratio = args.validation_ratio
+
+    save_dir = args.save_path
+    os.makedirs(save_dir, exist_ok=True)
+
     atoms_train = []
     atoms_validation = []
     atoms_test = []
 
-    random.seed(42)
+    random.seed(args.seed)
 
     for vasp_file in vasp_files:
         atoms_list = AtomsAdaptor.from_file(filename=vasp_file)
@@ -54,10 +48,31 @@ def main():
 
     # Save the training, validation, and test datasets to xyz files
 
-    write(f"{save_dir}/train.xyz", atoms_train)
-    write(f"{save_dir}/valid.xyz", atoms_validation)
-    write(f"{save_dir}/test.xyz", atoms_test)
+    write(f"{save_dir}/train.xyz", atoms_train, format="extxyz")
+    write(f"{save_dir}/valid.xyz", atoms_validation, format="extxyz")
+    write(f"{save_dir}/test.xyz", atoms_test, format="extxyz")
 
 
 if __name__ == "__main__":
-    main()
+    # Some important arguments
+    parser = argparse.ArgumentParser()
+
+    # path parameters
+    parser.add_argument("--data_path", type=str, default=None, help="vasprun data path")
+    parser.add_argument("--train_ratio", type=float, default=0.8, help="train ratio")
+    parser.add_argument(
+        "--validation_ratio", type=float, default=0.1, help="validation ratio"
+    )
+    parser.add_argument(
+        "--save_path",
+        type=str,
+        default="./xyz_files",
+        help="path to save the xyz files",
+    )
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=42,
+    )
+    args = parser.parse_args()
+    main(args)