Input should be a valid dictionary or instance of MergeConfiguration

[Hugo-Calero]

Hello!

I am trying to use the following notebook (https://github.com/arcee-ai/mergekit/blob/main/notebook.ipynb) with this yaml configuration file:

'''
slices:

• sources:
  • model: HuggingFaceTB/SmolLM-360M-Instruct
    layer_range: [0, 2]
• sources:
  • model: HuggingFaceTB/SmolLM-360M-Instruct
    layer_range: [11,32]

merge_method: passthrough
dtype: bfloat16
'''

I am getting the following error when running the code. Does merge kit require a specific version of pydantic? The one I have installed is 2.7.1
If anyone experienced anything similar to this or knows how to solve it, any help would be greatly appreciated :)

Error:
Traceback (most recent call last):
File "/Users/hugocalero/Desktop/PruneMe/slice_with_mergekit/merge_me.py", line 18, in
run_merge(
File "/Users/hugocalero/Desktop/PruneMe/slice_with_mergekit/mergekit/mergekit/merge.py", line 87, in run_merge
).plan_to_disk(out_path=out_path)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/hugocalero/Desktop/PruneMe/slice_with_mergekit/mergekit/mergekit/plan.py", line 259, in plan_to_disk
self._plan()
File "/Users/hugocalero/Desktop/PruneMe/slice_with_mergekit/mergekit/mergekit/plan.py", line 308, in _plan
ConfigReader(
File "/Users/hugocalero/Desktop/PruneMe/venv/lib/python3.11/site-packages/pydantic/main.py", line 176, in init
self.pydantic_validator.validate_python(data, self_instance=self)
pydantic_core.pydantic_core.ValidationError: 1 validation error for ConfigReader
config
Input should be a valid dictionary or instance of MergeConfiguration [type=model_type, input_value=MergeConfiguration(merge...te=None, out_dtype=None), input_type=MergeConfiguration]
For further information visit https://errors.pydantic.dev/2.7/v/model_type

[metric-space]

Hey @Hugo-Calero
The following seems to work for me

slices:
  - sources:
    - model: HuggingFaceTB/SmolLM-360M-Instruct
      layer_range: [0, 2]
    - model: HuggingFaceTB/SmolLM-360M-Instruct
      layer_range: [11,32]

merge_method: passthrough
dtype: bfloat16

Is it possible the space between slices and sources in your above given config file is the problem?

[Hugo-Calero]

Hello,

Thank you for your answer. I have tried deleting such space, but I am still getting the same error saying the input is not a valid dict or instance of MergeConfiguration.

The merge_config variable that is loaded is:
merge_method='passthrough' slices=[OutputSliceDefinition(sources=[InputSliceDefinition(model=ModelReference(model=ModelPath(path='HuggingFaceTB/SmolLM-360M-Instruct', revision=None), lora=None, override_architecture=None), layer_range=(0, 2), parameters=None), InputSliceDefinition(model=ModelReference(model=ModelPath(path='HuggingFaceTB/SmolLM-360M-Instruct', revision=None), lora=None, override_architecture=None), layer_range=(11, 32), parameters=None)], base_model=None, residual_weight=None, parameters=None)] models=None parameters=None base_model=None dtype='bfloat16' tokenizer_source=None tokenizer=None chat_template=None out_dtype=None