[BUG] EvaluatorArguments,

[Harshit-tech9]

Description:

I encountered an issue while running the script feedback_training_raft_t2t.py. The script fails with the following traceback:

Traceback (most recent call last):
File "/hdd/harshit-space/Feedbacknmt/train_scripts/../src/feedback_training_raft_t2t.py", line 15, in
from lmflow.args import (
File "/home/eztech/.pyenv/versions/feedbackmt/lib/python3.10/site-packages/lmflow/args.py", line 533, in
"evaluator": EvaluatorArguments,
NameError: name 'EvaluatorArguments' is not defined

Details:

The error occurs in the file lmflow/args.py at line 533, where EvaluatorArguments is referenced.
It seems that EvaluatorArguments is not defined or imported in lmflow/args.py, causing a NameError.

Expected Behavior:

The script should run without errors related to undefined names.

Environment:

Python version: 3.10
lmflow version: (Include the version of lmflow you're using)
Operating System: (Include your OS and version)
Additional Information:

Please let me know if there are any additional details required. I would appreciate any guidance or fixes for this issue.

[research4pan]

Thanks for your suggestions! For RAFT scripts, the LMFlow version needs to be very low, such as v0.0.4. You may try downgrading LMFlow to see if the problem is resolved. We are currently working on integrating supports for higher-version supports for RAFT, and will get back to you once it is available. Thanks for your understanding 🙏