Hi. Thank you.
The last line ensures compatibility with the mDeBERTa model.
Yeah, that’s what I had done, otherwise the model would crash.
This is my docker-compose.yml:
ner-demo:
container_name: ner-demo
image: deeppavlov/deeppavlov
environment:
- CONFIG=ner_demo_mdeberta_address
ports:
- "127.0.0.1:5000:5000"
volumes:
- ./deeppavlov/ner_demo_mdeberta_address.json:/usr/local/lib/python3.10/site-packages/deeppavlov/configs/classifiers/ner_demo_mdeberta_address.json
- ./data:/root/.deeppavlov
- ./venv:/venv
entrypoint:
- /bin/sh
- -c
- |
/usr/local/bin/python3.10 -m pip install sentencepiece==0.2.0 protobuf==3.20
python -m deeppavlov riseapi ner_demo_mdeberta_address -p 5000 -d
I put 0.2.0 because that’s what was in the requirements for that model. Without the version specified, it installs the same anyway.
Even if I make it the same as your commands:
volumes:
- ./deeppavlov/ner_demo_mdeberta_address.json:/usr/local/lib/python3.10/site-packages/deeppavlov/configs/classifiers/ner_demo_mdeberta_address.json
- ./data:/root/.deeppavlov
- ./venv:/venv
entrypoint:
- /bin/sh
- -c
- |
python -m deeppavlov install ner_demo_mdeberta_address
python -m pip install sentencepiece protobuf==3.20
python -m deeppavlov riseapi ner_demo_mdeberta_address -p 5000 -d
I get the same results.
–
Could these warnings give a hint about why I get the differences?
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
/usr/local/lib/python3.10/site-packages/transformers/convert_slow_tokenizer.py:454: UserWarning: The sentencepiece tokenizer that you are converting to a fast tokenizer uses the byte fallback option which is not implemented in the fast tokenizers. In practice this means that the fast version of the tokenizer can produce unknown tokens whereas the sentencepiece version would have converted these unknown tokens into a sequence of byte tokens matching the original piece of text.
warnings.warn(
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
(takes a long time, and then…)
Some weights of the model checkpoint at microsoft/mdeberta-v3-base were not used when initializing DebertaV2ForTokenClassification: [‘deberta.embeddings.word_embeddings._weight’, ‘lm_predictions.lm_head.LayerNorm.weight’, ‘lm_predictions.lm_head.LayerNorm.bias’, ‘mask_predictions.classifier.bias’, ‘mask_predictions.dense.bias’, ‘deberta.embeddings.position_embeddings.weight’, ‘mask_predictions.dense.weight’, ‘lm_predictions.lm_head.bias’, ‘mask_predictions.LayerNorm.weight’, ‘lm_predictions.lm_head.dense.weight’, ‘mask_predictions.LayerNorm.bias’, ‘mask_predictions.classifier.weight’, ‘deberta.embeddings.position_embeddings._weight’, ‘lm_predictions.lm_head.dense.bias’]
- This IS expected if you are initializing DebertaV2ForTokenClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing DebertaV2ForTokenClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of DebertaV2ForTokenClassification were not initialized from the model checkpoint at microsoft/mdeberta-v3-base and are newly initialized: [‘classifier.bias’, ‘classifier.weight’]
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
Many thanks!
