Closed mjpost closed 2 years ago
On a related question, how does this dictionary map to the list of models in https://unbabel.github.io/COMET/html/models.html#faq
{'emnlp20-comet-rank': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt20/emnlp20-comet-rank.tar.gz',
'wmt20-comet-da': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt20/wmt20-comet-da.tar.gz',
'wmt20-comet-qe-da': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt20/wmt20-comet-qe-da.tar.gz',
'wmt21-comet-mqm': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt21/wmt21-comet-mqm.tar.gz',
'wmt21-cometinho-da': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt21/wmt21-cometinho-da.tar.gz',
'wmt21-comet-qe-mqm': 'https://unbabel-experimental-models.s3.amazonaws.com/comet/wmt21/wmt21-comet-qe-mqm.tar.gz'
}
@alvations that's something I have to fix a long time ago already! Documentation is extremely outdated...
wmt-large-da-estimator-1719 | wmt20-comet-da |
---|---|
↑wmt-base-da-estimator-1719 | not supported in v1.0 |
↑wmt-large-qe-estimator-1719 | wmt20-comet-qe-da |
↓wmt-large-hter-estimator | not supported in v1.0 |
↓wmt-base-hter-estimator | not supported in v1.0 |
↑emnlp-base-da-ranker | emnlp20-comet-rank |
All other models are from our 2021 paper and are not in that documentation!
Regarding @mjpost request, it does print the available metrics when you pass an invalid model name:
comet-score -s src.de -t hyp1.en -r ref.en --gpus 0 --model test
usage: comet-score [-h] [-s SOURCES] [-t TRANSLATIONS [TRANSLATIONS ...]] [-r REFERENCES] [-d SACREBLEU_DATASET] [--batch_size BATCH_SIZE] [--gpus GPUS] [--accelerator {dp,ddp}] [--to_json TO_JSON]
[--model MODEL] [--model_storage_path MODEL_STORAGE_PATH] [--mc_dropout MC_DROPOUT] [--seed_everything SEED_EVERYTHING] [--num_workers NUM_WORKERS] [--disable_bar] [--disable_cache]
[--disable_length_batching] [--print_cache_info]
comet-score: error: test is not a valid checkpoint path or model choice. Choose from dict_keys(['emnlp20-comet-rank', 'wmt20-comet-da', 'wmt20-comet-qe-da', 'wmt21-comet-mqm', 'wmt21-cometinho-da', 'wmt21-comet-qe-mqm'])
When I pass --model wmt21-comet-qe-da
Global seed set to 12
Downloading refless-wmt21-comet-da.tar.gz
refless-wmt21-comet-da.tar.gz: 0.00B [00:00, ?B/s]
Traceback (most recent call last):
File "comet/cli/score.py", line 327, in
which means the model name (taken from doc) is not wrong according to your post, but not able to download. I'll use wmt20-comet-qe-da instead
you are right @vince62s! The link is broken for that model. I'll create a new issue for this
🚀 Feature
COMET should output a list of available models if
-m
is used with an invalid model name.Motivation
It's a bit of a pain to figure out the available models from the CLI. I had to come to the Github page.
Alternatives
Additional context