[rag] missing a working End-to-end evaluation example

[stas00]

I'm going to try to write tests for examples/rag (#7715), but first I'm trying to figure out how it works.

Would it be possible to add a full End-to-end evaluation invocation example in https://github.com/huggingface/transformers/blob/master/examples/rag/README.md#end-to-end-evaluation? i.e. with the correct data.

I tested https://github.com/huggingface/transformers/blob/master/examples/rag/README.md#retrieval-evaluation and it worked, but if I try to adapt the same params for e2e it crashes with:

$ python eval_rag.py --model_name_or_path facebook/rag-sequence-nq --model_type rag_sequence \
--evaluation_set output/biencoder-nq-dev.questions --gold_data_path output/biencoder-nq-dev.pages \
--predictions_path output/retrieval_preds.tsv --eval_mode e2e --gold_data_mode qa --n_docs 5 \
--print_predictions
2020-11-03 22:07:33.124277: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
INFO:__main__:Evaluate the following checkpoints: ['facebook/rag-sequence-nq']
INFO:__main__:Calculating metrics based on an existing predictions file: output/retrieval_preds.tsv
Traceback (most recent call last):
  File "eval_rag.py", line 314, in <module>
    main(args)
  File "eval_rag.py", line 280, in main
    score_fn(args, args.predictions_path, args.gold_data_path)
  File "eval_rag.py", line 46, in get_scores
    data = pd.read_csv(gold_data_path, sep="\t", header=None)
  File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/pandas/io/parsers.py", line 686, in read_csv
    return _read(filepath_or_buffer, kwds)
  File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/pandas/io/parsers.py", line 458, in _read
    data = parser.read(nrows)
  File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/pandas/io/parsers.py", line 1196, in read
    ret = self._engine.read(nrows)
  File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/pandas/io/parsers.py", line 2155, in read
    data = self._reader.read(nrows)
  File "pandas/_libs/parsers.pyx", line 847, in pandas._libs.parsers.TextReader.read
  File "pandas/_libs/parsers.pyx", line 862, in pandas._libs.parsers.TextReader._read_low_memory
  File "pandas/_libs/parsers.pyx", line 918, in pandas._libs.parsers.TextReader._read_rows
  File "pandas/_libs/parsers.pyx", line 905, in pandas._libs.parsers.TextReader._tokenize_rows
  File "pandas/_libs/parsers.pyx", line 2042, in pandas._libs.parsers.raise_parser_error
pandas.errors.ParserError: Error tokenizing data. C error: Expected 5 fields in line 2, saw 6

I think it needs a different input data.

And we need 2 functional examples: for qa and ans each.

I can handle adding this to the doc if you tell me what to add.

Thanks.

@patrickvonplaten, @lhoestq

[shamanez]

@stas00

Can you please write a test code for finetune.sh.

[stas00]

As you can see I'm waiting for this ticket to be addressed before I'm able to write the tests.

Perhaps you can address that, and then I will have all the info needed to write the tests.

[stas00]

Until then please file a normal issue about it. I haven't done any rag work yet, so that's why I'm asking for support.

[patrickvonplaten]

@lhoestq is working on this at the moment :-)

[lhoestq]

Actually I'm working on the finetuning script example, not eval ;)
But maybe this can help with adding a test for the eval script example.

[stale]

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.

[stas00]

stale