The setting is
# Set default values for method and router
method="ProxyFineTuning"
router="normal"
# Define the model size lists
models=("7B" "14B")
# Define the datasets to iterate over
datasets=("gsm8k")
small_ft_model_path="${base_path}-${small_model}-Chat"
Check specific key in slm_ft_kwargs if needed
slm_ft_kwargs only receive stop_word_ids but no position_ids.
slm_ft_kwargs: {'stop_word_ids': [[151645], [151644], [1], [2]]}
position_ids not found in slm_ft_kwargs
and prepare_inputs_for_generation returns AttributeError: 'NoneType' object has no attribute 'shape'