Efficient Attentions for Long Document Summarization https://arxiv.org/abs/2104.02112 Note: The average length of the reference summaries is ~3,000 characters, or ~600 tokens as tokenized by GPT-NeoX. For causal models, it is recommended to set `max_gen_toks` sufficently large (e.g.
| 419 | |
| 420 | |
| 421 | class GovReport(_SCROLLSSummaryTask): |
| 422 | """Efficient Attentions for Long Document Summarization |
| 423 | https://arxiv.org/abs/2104.02112 |
| 424 | |
| 425 | Note: The average length of the reference summaries is ~3,000 |
| 426 | characters, or ~600 tokens as tokenized by GPT-NeoX. For causal models, |
| 427 | it is recommended to set `max_gen_toks` sufficently large (e.g. 1024) |
| 428 | to allow a full summary to be generated. |
| 429 | """ |
| 430 | |
| 431 | DATASET_NAME = "gov_report" |
| 432 | |
| 433 | |
| 434 | class SummScreenFD(_SCROLLSSummaryTask): |
nothing calls this directly
no outgoing calls
no test coverage detected