search

Riccorl / transformer-srl

Reimplementation of a BERT based model (Shi et al, 2019), currently the state-of-the-art for English SRL. This model implements also predicate disambiguation.
69 stars 9 forks source link

RuntimeError: Error loading state dict for TransformerSrlSpan #10

Closed creatorrr closed 3 years ago

creatorrr commented 3 years ago

On running the example in README.md, getting the following error:

RuntimeError: Error loading state dict for TransformerSrlSpan
        Missing keys: ['transformer.embeddings.position_ids']
        Unexpected keys: []

(Model downloaded from the Dropbox link mentioned)

creatorrr commented 3 years ago

Environment:

python = "^3.8"
transformer-srl = "^2.5"
spacy = "2.3"
Riccorl commented 3 years ago

Can you share all the packages in the env?

I run it recently, and it worked fine in a clean environment.

creatorrr commented 3 years ago

That’s weird. I used it in a fresh environment too. I’ll do it once more and report but what python version were you using?

creatorrr commented 3 years ago

No luck. :-/

Same error using python3.7

Stacktrace:

Python 3.7.5 (default, Feb 23 2021, 13:22:40)
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from transformer_srl import dataset_readers, models, predictors

>>>
>>> predictor = predictors.SrlTransformersPredictor.from_path("./srl_bert_base_conll2012.tar.gz", "transformer_srl")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/transformer_srl/predictors.py", line 159, in from_path
    load_archive(archive_path, cuda_device=cuda_device),
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.7/site-packages/allennlp/models/archival.py", line 208, in load_archive
    model = _load_model(config.duplicate(), weights_path, serialization_dir, cuda_device)
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.7/site-packages/allennlp/models/archival.py", line 246, in _load_model
    cuda_device=cuda_device,
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.7/site-packages/allennlp/models/model.py", line 406, in load
    return model_class._load(config, serialization_dir, weights_file, cuda_device)
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.7/site-packages/allennlp/models/model.py", line 349, in _load
    f"Error loading state dict for {model.__class__.__name__}\n\t"
RuntimeError: Error loading state dict for TransformerSrlSpan
        Missing keys: ['transformer.embeddings.position_ids']
        Unexpected keys: []
Riccorl commented 3 years ago

Last time i tried python 3.6 Someone opened a similar issue #7, but I tried in a new env and it worked (https://github.com/Riccorl/transformer-srl/issues/7#issuecomment-810530073).

Try with

conda create -n srl-test python=3.6
conda activate srl-test
pip install transformer-srl==2.4.6
echo '{"sentence": "Did Uriah honestly think he could beat the game in under three hours?"}' | \
allennlp predict path/to/srl_bert_base_conll2012.tar.gz - --include-package transformer_srl

It should work

creatorrr commented 3 years ago

Previously, I had installed transformer-srl==2.5 (the latest one pypi). These instructions (with 2.4.6) led to new errors however. Seems like they depend on an older version of allennlp.

Stacktrace:

Traceback (most recent call last):
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/bin/allennlp", line 8, in <module>
    sys.exit(run())
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.6/site-packages/allennlp/__main__.py", line 34, in run
    main(prog="allennlp")
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.6/site-packages/allennlp/commands/__init__.py", line 117, in main
    import_module_and_submodules(package_name)
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.6/site-packages/allennlp/common/util.py", line 354, in import_module_and_submodules
    import_module_and_submodules(subpackage)
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/.venv/lib/python3.6/site-packages/allennlp/common/util.py", line 343, in import_module_and_submodules
    module = importlib.import_module(package_name)
  File "/usr/lib/python3.6/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 955, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 665, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 678, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "/home/ubuntu/src/github.com/Riccorl/transformer-srl/transformer_srl/predictors.py", line 3, in <module>
    from allennlp.data.tokenizers.token_class import Token
ModuleNotFoundError: No module named 'allennlp.data.tokenizers.token_class'
Riccorl commented 3 years ago

Did you install with --upgrade option? Because AllenNLP got a major update between 2.4.6 (the one that works with the pretrained model) and the most recent one

creatorrr commented 3 years ago

Yep, same problem. For v2.5, could it be an issue with the dropbox archive?

creatorrr commented 3 years ago

Same with 2.4.11

Riccorl commented 3 years ago

The pretrained weights on Dropbox only work with 2.4.6

creatorrr commented 3 years ago

ah, gotcha. could you try to reproduce once more when you get a chance? Not sure what I am doing wrong, followed the instructions in your comment to the t. :-|

Riccorl commented 3 years ago

Just tried the following instructions on an Ubuntu machine

wget https://www.dropbox.com/s/4tes6ypf2do0feb/srl_bert_base_conll2012.tar.gz
conda create -n srl-test python=3.6
conda activate srl-test
pip install transformer-srl==2.4.6
echo '{"sentence": "Did Uriah honestly think he could beat the game in under three hours?"}' | \
allennlp predict srl_bert_base_conll2012.tar.gz - --include-package transformer_srl

and I got the correct output

input 0:  {"sentence": "Did Uriah honestly think he could beat the game in under three hours?"}
prediction:  {"verbs": [{"verb": "Did", "description": "[do.01: Did] Uriah honestly think he could beat the game in under three hours ?", "tags": ["B-V", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O"], "frame": "do.01", "frame_score": 0.9999996423721313, "lemma": "do"}, {"verb": "think", "description": "Did [ARG0: Uriah] [ARGM-ADV: honestly] [think.01: think] [ARG1: he could beat the game in under three hours] ?", "tags": ["O", "B-ARG0", "B-ARGM-ADV", "B-V", "B-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "I-ARG1", "O"], "frame": "think.01", "frame_score": 1.0, "lemma": "think"}, {"verb": "could", "description": "Did Uriah honestly think he [go.04: could] beat the game in under three hours ?", "tags": ["O", "O", "O", "O", "O", "B-V", "O", "O", "O", "O", "O", "O", "O", "O"], "frame": "go.04", "frame_score": 0.10186540335416794, "lemma": "could"}, {"verb": "beat", "description": "Did Uriah honestly think [ARG0: he] [ARGM-MOD: could] [beat.03: beat] [ARG1: the game] [ARGM-TMP: in under three hours] ?", "tags": ["O", "O", "O", "O", "B-ARG0", "B-ARGM-MOD", "B-V", "B-ARG1", "I-ARG1", "B-ARGM-TMP", "I-ARGM-TMP", "I-ARGM-TMP", "I-ARGM-TMP", "O"], "frame": "beat.03", "frame_score": 0.9999936819076538, "lemma": "beat"}], "words": ["Did", "Uriah", "honestly", "think", "he", "could", "beat", "the", "game", "in", "under", "three", "hours", "?"]}
creatorrr commented 3 years ago

Worked this time for some reason. 🤷 Thanks so much for patiently helping! Maybe update the README to install 2.4.6 ?

Riccorl commented 3 years ago

Worked this time for some reason. 🤷 Thanks so much for patiently helping! Maybe update the README to install 2.4.6 ?

Glad I could help :)

Yeah maybe you are right, I should write better to use 2.4.6