GitLab

[WIP] Upstream `hf-causal` and `hf-seq2seq` model implementations

Enable "low_cpu_mem_usage" to reduce the memory usage of HF models

hotfix(gpt2): Remove vocab-size logits slice

* add accuracy metric to crows-pairs

* Added CrowSPairs

* Fixed Bug

* Changed to absolute Value

* Docstring Changes

* Update crowspairs.py

Finegrained Tasks

* Update __init__.py

* Finegrained Tasks

* Add `crows_pairs` version metadata

* added toxigen task

* removed previous doc_to_text

* Add train split for few-shot samples

* Add task version metadata

* use pile links hosted on the eye

* update links in dataset_infos

Update `lambada_openai` multilingual data source

Bump `triviaqa` version

* Update `actions/setup-python`in  CI workflows

Replace stale `triviaqa` dataset link

Cleanup `README.md` and package deps

Add the original `LAMBADA` dataset

Remove unsupported `gitlab` link in pre-commit config

This reverts commit 63bdb4b0.

* Add `TextSynth` API

Remove custom datasets that are in HF

[deps] Use minimum versioning for `numexpr`

Manually concat tokenizer revision with subfolder