[Bug]: convert .safetensors z base image to diffusers

[BenDes21]

What happened?

Hello there, I have a hard time for convert my .safetensors finetuned z base image model to transformers then use it for train my lora on it with onetrainer, Im getting File "transformers/tokenization_utils_base.py", line 1210, in _set_model_specific_special_tokens
self.SPECIAL_TOKENS_ATTRIBUTES = self.SPECIAL_TOKENS_ATTRIBUTES + list(special_tokens.keys())
AttributeError: 'list' object has no attribute 'keys' , Is anyone know how to fix the issue ? 🙂 Not sure its a bug Thanks

What did you expect would happen?

.safetensors from full finetuned z base image supposed to be in diffusers files trainable with onetrainer ( lora training )

Relevant log output

Generate and upload debug_report.log

No response

[O-J1]

.safetensors from full finetuned z base image supposed to be in diffusers files trainable with onetrainer ( lora training )

Based on what you implied your model did not come from OT, so not an OT bug.

Please don’t submit help requests as bugs, you’ve not provided debug report log either?

If a model trained in OneTrainer fails to load and you can provide the steps or I have misunderstood then please clarify.

[BenDes21]

.safetensors from full finetuned z base image supposed to be in diffusers files trainable with onetrainer ( lora training )

Based on what you implied your model did not come from OT, so not an OT bug.

Please don’t submit help requests as bugs, you’ve not provided debug report log either?

If a model trained in OneTrainer fails to load and you can provide the steps or I have misunderstood then please clarify.

Hi there, sorry for the wrong issue type, Im not sure about which one use for simply ask help.

Can you confirm that a model full finetuned on another trainer than OT ( let's said musubi ) cannot be used ( when converted to transformers files ) for training lora's on it ?

[dxqb]

converting to discussion
a full finetune can be loaded by OneTrainer, when as .safetensors or as diffusers

[BenDes21]

Hi, Im using onetrainer cli and getting this error when using .safetensors or transformers files :

[OneTrainer] + python scripts/train.py --config-path /workspace/OneTrainer/ero_forked_1024_rank64.json
Could not set centered_wd_mode as CenteredWDMode.FLOAT8
Clearing cache directory ! You can disable this if you want to continue using the same cache.
No backup found, continuing without backup...
Traceback (most recent call last):
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 167, in load
    self.__load_internal(
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 51, in __load_internal
    raise Exception("not an internal model")
Exception: not an internal model

Traceback (most recent call last):
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/utils/hub.py", line 479, in cached_files
    hf_hub_download(
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 106, in _inner_fn
    validate_repo_id(arg_value)
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 154, in validate_repo_id
    raise HFValidationError(
huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/workspace/output/Ero_v2-000036-ComfyUI.safetensors'. Use `repo_type` argument if needed.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/tokenization_utils_base.py", line 1994, in from_pretrained
    resolved_config_file = cached_file(
                           ^^^^^^^^^^^^
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/utils/hub.py", line 322, in cached_file
    file = cached_files(path_or_repo_id=path_or_repo_id, filenames=[filename], **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/utils/hub.py", line 532, in cached_files
    _get_cache_file_to_return(path_or_repo_id, filename, cache_dir, revision, repo_type)
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/utils/hub.py", line 143, in _get_cache_file_to_return
    resolved_file = try_to_load_from_cache(
                    ^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 106, in _inner_fn
    validate_repo_id(arg_value)
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 154, in validate_repo_id
    raise HFValidationError(
huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/workspace/output/Ero_v2-000036-ComfyUI.safetensors'. Use `repo_type` argument if needed.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 175, in load
    self.__load_diffusers(
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 76, in __load_diffusers
    tokenizer = Qwen2Tokenizer.from_pretrained(
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspace/OneTrainer/venv/lib/python3.12/site-packages/transformers/tokenization_utils_base.py", line 2014, in from_pretrained
    raise OSError(
OSError: Can't load tokenizer for '/workspace/output/Ero_v2-000036-ComfyUI.safetensors'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure '/workspace/output/Ero_v2-000036-ComfyUI.safetensors' is the correct path to a directory containing all relevant files for a Qwen2Tokenizer tokenizer.

Traceback (most recent call last):
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 183, in load
    self.__load_safetensors(
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 154, in __load_safetensors
    raise NotImplementedError("Loading of single file Z-Image models not supported. Use the diffusers model instead. Optionally, transformer-only safetensor files can be loaded by overriding the transformer.")
NotImplementedError: Loading of single file Z-Image models not supported. Use the diffusers model instead. Optionally, transformer-only safetensor files can be loaded by overriding the transformer.

Traceback (most recent call last):
  File "/workspace/OneTrainer/scripts/train.py", line 47, in <module>
    main()
  File "/workspace/OneTrainer/scripts/train.py", line 34, in main
    trainer.start()
  File "/workspace/OneTrainer/modules/trainer/GenericTrainer.py", line 129, in start
    self.model = self.model_loader.load(
                 ^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspace/OneTrainer/modules/modelLoader/GenericLoRAModelLoader.py", line 51, in load
    base_model_loader.load(model, model_type, model_names, weight_dtypes, quantization)
  File "/workspace/OneTrainer/modules/modelLoader/ZImageModelLoader.py", line 192, in load
    raise Exception("could not load model: " + model_names.base_model)
Exception: could not load model: /workspace/output/Ero_v2-000036-ComfyUI.safetensors
/workspace/OneTrainer/venv/lib/python3.12/site-packages/tensorboard/default.py:30: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  import pkg_resources
root@bc1f7f4b1f75:/workspace/OneTrainer# TensorFlow installation not found - running with reduced feature set.

NOTE: Using experimental fast data loading logic. To disable, pass
    "--load_fast=false" and report issues on GitHub. More details:
    https://github.com/tensorflow/tensorboard/issues/4784

Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all
TensorBoard 2.20.0 at http://localhost:6006/ (Press CTRL+C to quit)
root@bc1f7f4b1f75:/workspace/OneTrainer# 

This model has been finetuned with musubi tuner and I set the path here in my config file : "base_model_name": "/workspace/output/Ero_v2-000036-ComfyUI.safetensors"

[dxqb]

use Override transformer, not base model. Your .safetensors file (likely) only contains the transformer, not the entire model. The base model must still be the diffusers repo.

[BenDes21]

thanks seems to work