a2t bug

[ko120]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Steps to reproduce the behavior:

1. Run following command textattack ...
2. Run following code ...
3. See error

Expected behavior
A clear and concise description of what you expected to happen.
It ask us to input truncate_words_to
Screenshots or Traceback
If applicable, add screenshots to help explain your problem. Also, copy and paste tracebacks produced by the bug.

System Information (please complete the following information):

• OS: [e.g. MacOS, Linux, Windows]
• Library versions (e.g. torch==1.7.0, transformers==3.3.0)
• Textattack version

Additional context
Add any other context about the problem here.

[qiyanjun]

@jinyongyoo mind to take a look?

[jinyongyoo]

@ko120 @qiyanjun

Looks like the issue is with the truncate_words_to keyword argument which isn't part of GreedyWordSwapWIR. The argument was add in PR #747. @qiyanjun Could you share the background behind the PR and why that argument might have been added?

[xcegin]

I have come across the same issue yesterday, suprised to see this hanging here from November. Any updates?

[qiyanjun]

Sorry for the delay.. will take a careful look

PR #747 , added a max length constraint in

search_method = GreedyWordSwapWIR(wir_method="gradient")

    max_len = getattr(model_wrapper, "max_length", None) or min(
        1024, model_wrapper.tokenizer.model_max_length, model_wrapper.model.config.max_position_embeddings - 2
    )
    search_method = GreedyWordSwapWIR(wir_method="gradient", truncate_words_to=max_len)

[xcegin]

In the GreedyWordSwapWIR class from the PR here, I dont see any argument truncate_words argument in the __init__ method of the class. Maybe an oversight (I am no contributor though)?

[qiyanjun]

Please feel free to submit a PR to fix this..Sent from my iPhoneOn Apr 25, 2024, at 12:16, JanCe ***@***.***> wrote: In the GreedyWordSwapWIR class from the PR here, I dont see any argument truncate_words argument in the __init__ method of the class. Maybe an oversight (I am no contributor though)? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***>

[qiyanjun]

Confirmed — GreedyWordSwapWIR.__init__ never declared truncate_words_to, even though a2t_yoo_2021.py has been passing it since PR #747 added the max-length guard. Opened #831, which adds the parameter and gives it real behavior: when set, it caps the search to the first N word indices, bounding cost on very long inputs (which was presumably #747's intent). Will close automatically once that merges.