Language I am using the model on (English, Chinese....): English
The problem arise when using:
[ ] the official example scripts: (give details)
I use script run_ner.py in order to finetune xlmroberta on conll03 dataset.
The script executed with no problems. But the file "config.json" in the output directory is not correct.
The tasks I am working on is:
[ ] an official GLUE/SQUaD task: (give the name)
Task: NER
Dataset: conll03
To Reproduce
Steps to reproduce the behavior:
1.
I run the script: run_ner.py as follows:
python run_ner.py --data_dir 0-data --model_type 'xlmroberta' --model_name_or_path 'xlm-roberta-large' --output_dir 1-out --max_seq_length 32 --do_train --do_eval --per_gpu_train_batch_size 8 --no_cuda --evaluate_during_training --logging_steps 1756 --save_steps 1756 --eval_all_checkpoints
Go to the output directory. The file "config.json" contains :
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
and
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
which are not expected in NER
🐛 Bug
Model I am using (Bert, XLNet....): xlmroberta
Language I am using the model on (English, Chinese....): English
The problem arise when using:
run_ner.py
in order to finetune xlmroberta on conll03 dataset.The script executed with no problems. But the file "config.json" in the output directory is not correct.
The tasks I am working on is:
To Reproduce
Steps to reproduce the behavior:
1. I run the script: run_ner.py as follows:
python run_ner.py --data_dir 0-data --model_type 'xlmroberta' --model_name_or_path 'xlm-roberta-large' --output_dir 1-out --max_seq_length 32 --do_train --do_eval --per_gpu_train_batch_size 8 --no_cuda --evaluate_during_training --logging_steps 1756 --save_steps 1756 --eval_all_checkpoints
Expected behavior
I expect that "config.json" contains something like: "id2label": { "0": "B-LOC", "1": "B-MISC", "2": "B-ORG", "3": "I-LOC", "4": "I-MISC", "5": "I-ORG", "6": "I-PER", "7": "O" }, and "label2id": { "B-LOC": 0, "B-MISC": 1, "B-ORG": 2, "I-LOC": 3, "I-MISC": 4, "I-ORG": 5, "I-PER": 6, "O": 7 },
Environment
Additional context