convert_weights.py

# 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.

"""Utility to convert weights to safetensors."""

import argparse

import torch

from .configuration_embed1 import CosmosEmbed1Config
from .modeling_embed1 import CosmosEmbed1


def parse_args():
    parser = argparse.ArgumentParser(description="Save model weights with optional format conversion and sharding.")
    parser.add_argument("--input_weights", type=str, required=True, help="Path to the input .pt weights file")
    parser.add_argument(
        "--output_weights",
        type=str,
        required=True,
        help="Path to the output directory where safetensors weights will be saved",
    )
    return parser.parse_args()


def main():
    args = parse_args()
    model = CosmosEmbed1(CosmosEmbed1Config()).to("cuda", dtype=torch.bfloat16)

    # remove tensor sharing
    model.qformer.cls.predictions.decoder.weight = torch.nn.Parameter(
        model.qformer.cls.predictions.decoder.weight.clone()
    )
    model.qformer.bert.embeddings.word_embeddings.weight = torch.nn.Parameter(
        model.qformer.bert.embeddings.word_embeddings.weight.clone()
    )
    model.qformer.cls.predictions.decoder.bias = torch.nn.Parameter(model.qformer.cls.predictions.decoder.bias.clone())
    model.qformer.cls.predictions.bias = torch.nn.Parameter(model.qformer.cls.predictions.bias.clone())

    with open(args.input_weights, "rb") as fp:
        state_dict = torch.load(fp)
    model.load_state_dict(state_dict, strict=True)

    model.save_pretrained(
        args.output_weights,
        safe_serialization=True,
        max_shard_size="500MB",
    )


if __name__ == "__main__":
    main()