Dimensions:
- level: token, sentence
- similarity measure:
- cosine
- MNLI
- NN
- Entail
- Entail-Contradictory
- LM:
- mpnet
- BART (not always)
- RoBERTa
- DeBERTa
- size: large, xlarge
- fine-tune strategy: basic (MLM), MNLI
- Sentence weighting:
- PageRank: Sum, Entropy
- pseudoreference:
- AnyRef
- Pegasus-CNNDM, Pegasus-Newsroom, Pegasus-large, Pegaus-xsum (no need to run/report)
- BART-CNNDM
- top-k
- top-p
- AnyRef
All experiment names must follow the order of dimension above, e.g., sentence-cosine-DeBERTa-large-MNLI-AnyRef-Pegaus-cnndm
- BERTScore-token
- RoBERTa-large, DeBERTa-large-mnli, DeBERTa-xlarge-mnli, DeBERTa-large (OOM)
- RoBERTa-large-topk, DeBERTa-large-mnli-topk, DeBERTa-xlarge-mnli-topk
- RoBERTa-large-topp, DeBERTa-large-mnli-topp, DeBERTa-xlarge-mnli-topp
- RoBERTa-large-anyref-{bart, pegasus-{xsum, cnndm, newsroom, large}}-{default, min, mean, constant-*}, DeBERTa (not run OOM)
- Why no IDF? Error? (please double check)
- BERTScore-sentence
- cosine:
- Vanilla: mpnet (done), RoBERTa-large (done), DeBERTa-large (done), DeBERTa-xlarge (not run)
- Add pseudoreference
- top-k (done), top-p (done)
- AnyRef (done)
- Add sentence weighting
- mpnet (done: entropy, sum), RoBERTa-large (done: entropy, sum)
- MNLI:
- Vanilla
- MNLI-{Bart, RoBERTa}-large-{NN, EO, EC}: done
- MNLI-DeBERta-xlarge-{NN, EO, EC}: done
- MNLI-DeBERta-large-{NN, EO, EC}: done
- Add sentence weighting
- MNLI-DeBERTa-xlarge-EC-{entropy, sum}: done
- MNLI-DeBERTa-large-EC-{entropy, sum}: done
- Add pseudorefence
- Nothing yet
- Vanilla
- cosine: