tsi-org/tango

1# coding=utf-82# Copyright 2023 The HuggingFace Inc. team.3#4# Licensed under the Apache License, Version 2.0 (the "License");5# you may not use this file except in compliance with the License.6# You may obtain a copy of the License at7#8#     http://www.apache.org/licenses/LICENSE-2.09#10# Unless required by applicable law or agreed to in writing, software11# distributed under the License is distributed on an "AS IS" BASIS,12# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.13# See the License for the specific language governing permissions and14# limitations under the License.15""" Conversion script for stable diffusion checkpoints which _only_ contain a contrlnet. """16 17import argparse18 19from diffusers.pipelines.stable_diffusion.convert_from_ckpt import download_controlnet_from_original_ckpt20 21 22if __name__ == "__main__":23    parser = argparse.ArgumentParser()24 25    parser.add_argument(26        "--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert."27    )28    parser.add_argument(29        "--original_config_file",30        type=str,31        required=True,32        help="The YAML config file corresponding to the original architecture.",33    )34    parser.add_argument(35        "--num_in_channels",36        default=None,37        type=int,38        help="The number of input channels. If `None` number of input channels will be automatically inferred.",39    )40    parser.add_argument(41        "--image_size",42        default=512,43        type=int,44        help=(45            "The image size that the model was trained on. Use 512 for Stable Diffusion v1.X and Stable Siffusion v2"46            " Base. Use 768 for Stable Diffusion v2."47        ),48    )49    parser.add_argument(50        "--extract_ema",51        action="store_true",52        help=(53            "Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights"54            " or not. Defaults to `False`. Add `--extract_ema` to extract the EMA weights. EMA weights usually yield"55            " higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning."56        ),57    )58    parser.add_argument(59        "--upcast_attention",60        action="store_true",61        help=(62            "Whether the attention computation should always be upcasted. This is necessary when running stable"63            " diffusion 2.1."64        ),65    )66    parser.add_argument(67        "--from_safetensors",68        action="store_true",69        help="If `--checkpoint_path` is in `safetensors` format, load checkpoint with safetensors instead of PyTorch.",70    )71    parser.add_argument(72        "--to_safetensors",73        action="store_true",74        help="Whether to store pipeline in safetensors format or not.",75    )76    parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output model.")77    parser.add_argument("--device", type=str, help="Device to use (e.g. cpu, cuda:0, cuda:1, etc.)")78    args = parser.parse_args()79 80    controlnet = download_controlnet_from_original_ckpt(81        checkpoint_path=args.checkpoint_path,82        original_config_file=args.original_config_file,83        image_size=args.image_size,84        extract_ema=args.extract_ema,85        num_in_channels=args.num_in_channels,86        upcast_attention=args.upcast_attention,87        from_safetensors=args.from_safetensors,88        device=args.device,89    )90 91    controlnet.save_pretrained(args.dump_path, safe_serialization=args.to_safetensors)92