TypeError: TrainingArguments.__init__() got an unexpected keyword argument 'finetune_mode'

[Zzz-zcy]

When I use the following code for LoRA fine-tuning, an error is reported. What could be the reason for this?
TypeError: TrainingArguments.init() got an unexpected keyword argument 'finetune_mode'.
finetuned_pipeline = pipeline.fit( inputs=train_data, prediction_length=1, finetune_mode="lora", learning_rate=1e-5, num_steps=500, batch_size=256, accelerator="npu" )

[shchur]

Please share the full stack trace for the error and the package version, otherwise it's impossible to diagnose the issue.

[Zzz-zcy]

chronos.version: 2.0.0

/home/....../chronos-forecasting_sft.py:130: FutureWarning: Fine-tuning support is experimental and may be changed in future versions.
finetuned_pipeline = pipeline.fit(
Traceback (most recent call last):
File "/home/....../chronos-forecasting_sft.py", line 130, in
finetuned_pipeline = pipeline.fit(
File "/home/ma-user/anaconda3/envs/PyTorch-2.1.0/lib/python3.10/site-packages/chronos/chronos2/pipeline.py", line 269, in fit
training_args = TrainingArguments(**training_kwargs)
TypeError: TrainingArguments.init() got an unexpected keyword argument 'finetune_mode'

[shchur]

You need to update to v2.2.0 with pip install -U chronos-forecasting, the argument wasn't present in v2.0.0.

[Zzz-zcy]

When fine-tuning Chronos-2 for covariate prediction tasks, there is a obvious issue of lagging predictions. How can this be resolved?