Add script to convert EMA network snapshots to inference checkpoints

Training writes EMA weights only to network-snapshot-*.pkl. This script
wraps the 'ema' module with batch_info and model_config from a
training-state checkpoint of the same run, so the EMA weights can be
loaded with cbottle.inference.load.

Signed-off-by: Peter Manshausen <pmanshausen@nvidia.com>

Author: Peter Manshausen

scripts/convert_ema_snapshot_to_checkpoint.py

@@ -0,0 +1,87 @@
+#!/usr/bin/env python3
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import pickle
+from pathlib import Path
+
+from cbottle import checkpointing
+
+
+def get_args():
+    parser = argparse.ArgumentParser(
+        description=(
+            "Convert a network-snapshot-*.pkl EMA artifact into a "
+            "training-state-style .checkpoint file for inference."
+        )
+    )
+    parser.add_argument(
+        "--snapshot",
+        type=str,
+        required=True,
+        help="Path to network-snapshot-*.pkl",
+    )
+    parser.add_argument(
+        "--reference-checkpoint",
+        type=str,
+        required=True,
+        help=(
+            "Path to a training-state-*.checkpoint from the same run. "
+            "Used to copy batch_info and model_config."
+        ),
+    )
+    parser.add_argument(
+        "--output",
+        type=str,
+        required=True,
+        help="Path to write converted EMA checkpoint.",
+    )
+    return parser.parse_args()
+
+
+def main():
+    args = get_args()
+
+    snapshot_path = Path(args.snapshot).expanduser().resolve()
+    reference_ckpt_path = Path(args.reference_checkpoint).expanduser().resolve()
+    output_path = Path(args.output).expanduser().resolve()
+
+    if not snapshot_path.is_file():
+        raise FileNotFoundError(str(snapshot_path))
+    if not reference_ckpt_path.is_file():
+        raise FileNotFoundError(str(reference_ckpt_path))
+
+    with open(snapshot_path, "rb") as f:
+        snapshot = pickle.load(f)
+    if "ema" not in snapshot:
+        raise KeyError(f"{snapshot_path} does not contain key 'ema'")
+    ema_model = snapshot["ema"]
+
+    with checkpointing.Checkpoint(str(reference_ckpt_path), "r") as ref_ckpt:
+        batch_info = ref_ckpt.read_batch_info()
+        model_config = ref_ckpt.read_model_config()
+
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    with checkpointing.Checkpoint(str(output_path), "w") as out_ckpt:
+        out_ckpt.write_model(ema_model)
+        out_ckpt.write_batch_info(batch_info)
+        out_ckpt.write_model_config(model_config)
+
+    print(f"Wrote EMA checkpoint: {output_path}")
+
+
+if __name__ == "__main__":
+    main()