I have been converting archival DreamBooths to LORA with the kohya_ss framework, for a month, and the conversions have all gone well.
However, I have been converting DreamBooth models trained prior to this period. Having now actually trained a couple of new DreamBooth models in the past week on the Colab, neither of them will convert. This is the error that results every time:
Describe the bug
I have been converting archival DreamBooths to LORA with the kohya_ss framework, for a month, and the conversions have all gone well.
However, I have been converting DreamBooth models trained prior to this period. Having now actually trained a couple of new DreamBooth models in the past week on the Colab, neither of them will convert. This is the error that results every time:
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮ │ C:\Users\[USER]\Desktop\KOHYA2\kohya_ss\networks\extract_lora_from_models.py:189 in <module> │ │ │ │ 186 parser = setup_parser() │ │ 187 │ │ 188 args = parser.parse_args() │ │ ❱ 189 svd(args) │ │ 190 │ │ │ │ C:\Users\[USER]\Desktop\KOHYA2\kohya_ss\networks\extract_lora_from_models.py:45 in svd │ │ │ │ 42 print(f"loading SD model : {args.model_org}") │ │ 43 text_encoder_o, _, unet_o = model_util.load_models_from_stable_diffusion_checkpoint(ar │ │ 44 print(f"loading SD model : {args.model_tuned}") │ │ ❱ 45 text_encoder_t, _, unet_t = model_util.load_models_from_stable_diffusion_checkpoint(ar │ │ 46 │ │ 47 # create LoRA network to extract weights: Use dim (rank) as alpha │ │ 48 if args.conv_dim is None: │ │ │ │ C:\Users\[USER]\Desktop\KOHYA2\kohya_ss\library\model_util.py:1059 in │ │ load_models_from_stable_diffusion_checkpoint │ │ │ │ 1056 │ │ │ torch_dtype="float32", │ │ 1057 │ │ ) │ │ 1058 │ │ text_model = CLIPTextModel._from_config(cfg) │ │ ❱ 1059 │ │ info = text_model.load_state_dict(converted_text_encoder_checkpoint) │ │ 1060 │ print("loading text encoder:", info) │ │ 1061 │ │ │ 1062 │ return text_model, vae, unet │ │ │ │ C:\Users\[USER]\Desktop\KOHYA2\kohya_ss\venv\lib\site-packages\torch\nn\modules\module.py:1604 │ │ in load_state_dict │ │ │ │ 1601 │ │ │ │ │ │ ', '.join('"{}"'.format(k) for k in missing_keys))) │ │ 1602 │ │ │ │ 1603 │ │ if len(error_msgs) > 0: │ │ ❱ 1604 │ │ │ raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( │ │ 1605 │ │ │ │ │ │ │ self.__class__.__name__, "\n\t".join(error_msgs))) │ │ 1606 │ │ return _IncompatibleKeys(missing_keys, unexpected_keys) │ │ 1607 │ ╰──────────────────────────────────────────────────────────────────────────────────────────────────╯If this can't be fixed, hopefully I could get access to an earlier version of the Colab.
Reproduction
N/A
Logs
No response
System Info
Colab standard