Transformers: run_ner.py load checkpoint issue

[giuliorav]

Hi,
I just launched the transformers/examples/run_ner.py script with my custom model:
python3 transformers/examples/run_ner.py --data_dir $INPUT_DATA_DIR \ --tokenizer_name $TOKENIZER_FILE_PATH --output_dir $OUTPUT_DIR --model_type camembert --labels $LABELS_DIR --model_name_or_path $BERT_MODEL --max_seq_length $MAX_LENGTH --num_train_epochs $NUM_EPOCHS --gradient_accumulation_steps $ACCUMULATION_STEPS --per_gpu_train_batch_size $BATCH_SIZE --save_steps $SAVE_STEPS --do_lower_case --do_train --do_eval --do_predict

Once the data for the train has been loaded, an error appear:
Traceback (most recent call last): File "transformers/examples/run_ner.py", line 567, in <module> main() File "transformers/examples/run_ner.py", line 496, in main global_step, tr_loss = train(args, train_dataset, model, tokenizer, labels, pad_token_label_id) File "transformers/examples/run_ner.py", line 132, in train global_step = int(args.model_name_or_path.split('-')[-1].split('/')[0]) ValueError: invalid literal for int() with base 10: 'pytorch_dump_folder'

Launching the same script a few hours ago the error did not appear, is it something related to the last updates #2134 ?

Thanks in advance.

[stefan-it]

Can confirm this issue -> a temporary workaround would be to change the line to:

if os.path.exists(args.model_name_or_path) and "checkpoint" in args.model_name_or_path:

[julien-c]

See also related (recent) fix on master: https://github.com/huggingface/transformers/commit/4d36472b96d144887cbe95b083f0d2091fd5ff03

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.