RecursionError: maximum recursion depth exceeded while getting the str of an object

[SMACY2017]

after i run ./scripts/run_chatbot.sh ./llama-hf-path/llama-7b-hf ./output_models/llama7b-lora-medical/ (same error when run app.py and other .sh files) , errors occur:

  File "/gpfs/share/home/2101212970/transformers-main/src/transformers/tokenization_utils_base.py", line 1141, in unk_token_id
    return self.convert_tokens_to_ids(self.unk_token)
  File "/gpfs/share/home/2101212970/transformers-main/src/transformers/tokenization_utils_fast.py", line 250, in convert_tokens_to_ids
    return self._convert_token_to_id_with_added_voc(tokens)
  File "/gpfs/share/home/2101212970/transformers-main/src/transformers/tokenization_utils_fast.py", line 260, in _convert_token_to_id_with_added_voc
    return self.unk_token_id
  File "/gpfs/share/home/2101212970/transformers-main/src/transformers/tokenization_utils_base.py", line 1141, in unk_token_id
    return self.convert_tokens_to_ids(self.unk_token)
  File "/gpfs/share/home/2101212970/transformers-main/src/transformers/tokenization_utils_base.py", line 1021, in unk_token
    return str(self._unk_token)
RecursionError: maximum recursion depth exceeded while getting the str of an object
[2023-04-10 12:35:05,027] [INFO] [launch.py:318:sigkill_handler] Killing subprocess 114376
[2023-04-10 12:35:05,028] [ERROR] [launch.py:324:sigkill_handler] ['/gpfs/share/home/2101212970/software/miniconda3/envs/lmflow/bin/python', '-u', 'examples/chatbot.py', '--local_rank=0', '--deepspeed', 'configs/ds_config_chatbot.json', '--model_name_or_path', './llama-hf-path/llama-7b-hf', '--lora_model_path', './output_models/llama7b-lora-medical/'] exits with return code = 1

if i make some changes like this:

import sys

sys.setrecursionlimit(10000)  # 增加最大递归深度限制

it will show
Segmentation fault (core dumped)

[SMACY2017]

In fact, I have changed the model file source of llama-7b-hf several times (through code conversion provided by LMflow or directly cloning llama-7b-hf on the huggingface), but this error is always reported

[research4pan]

Thanks for your interest in LMFlow! Looks like the downloaded ./llama-hf-path/llama-7b-hf has some bad attributes (or missing attributes). Could you please try ./scripts/run_chatbot.sh pinkmanlove/llama-7b-hf ./output_models/llama7b-lora-medical/ and see if the error occurs again? Thanks 😄

[SMACY2017]

Certainly. In fact, due to network reasons, I am unable to access the huggingface on the server, so I manually download it locally and then upload it to the server. I not only tried the pinkmanlove/llama-7b-hf you mentioned, but also tried other llama-7b-hf on the hugging face, still reporting this error

[SMACY2017]

I really don't know where the problem lies. I re-cloned LMFlow, reconfigured the virtual environment, and used the previously downloaded model weight files, which made the issue disappear. I guess there was a problem when I initially set up the environment. Since my server's access to GitHub is not very smooth, I manually operated the installation commands from GitHub for a few items in requirements.txt. Among them, "transformers @ git+https://github.com/huggingface/transformers@c612628045822f909020f7eb6784c79700813eda" might have had an issue. I got a 404 error when I directly accessed this URL, so I manually downloaded "https://github.com/huggingface/transformers#c612628045822f909020f7eb6784c79700813eda" (I replaced @ with # as advised by GPT-4). This might be the reason for the error. During the process of starting over, I blindly repeated the execution of pip install git+https://github.com/huggingface/transformers@c612628045822f909020f7eb6784c79700813eda until one of the attempts to access the network was successful. It was not easy at all! Thanks！

[research4pan]

Thanks for your efforts! Yeah, servers with limited access to github will face this kind of problems. I am glad to hear that you finally solve it. For package transformers, you may try cloning the original repository and then check out to the branch

git clone https://github.com/huggingface/transformers.git
cd transformers
git checkout c612628045822f909020f7eb6784c79700813eda
pip install -e .

You can also download the zip in a server that can access github via directly accessing the github page of that specific commit, then click code button and choose "Download ZIP". You may then transfer this ZIP file to your server.

Hope that can be some help and make things easy for you next time. Thanks very much for your efforts! 👍

[qiguanqiang]

my transformers version is 4.29.0, same problem. And I subtitute it with your version. Problem solved.

[shizhediao]

Great. Seems that this issue is solved. Closed.

[seanxuu]

Thanks for your efforts! Yeah, servers with limited access to github will face this kind of problems. I am glad to hear that you finally solve it. For package transformers, you may try cloning the original repository and then check out to the branch

git clone https://github.com/huggingface/transformers.git
cd transformers
git checkout c612628045822f909020f7eb6784c79700813eda
pip install -e .

You can also download the zip in a server that can access github via directly accessing the github page of that specific commit, then click code button and choose "Download ZIP". You may then transfer this ZIP file to your server.

Hope that can be some help and make things easy for you next time. Thanks very much for your efforts! 👍

It works!