Datasets: NeurIPS LLM Challenge 2023
Datasets that were under consideration for usage in my submission to the 2023 NeurIPS Large Language Model Efficiency Challenge.
Viewer • Updated • 3.4k • 32Note Ultimately used in my full eval submission, with exclusion of dolly_hhrlhf. Included only in Mistral-7B-sft-v1.
databricks/databricks-dolly-15k
Viewer • Updated • 28k • 655Note Used both for Mistral-7B-sft-v0 and Mistral-7B-sft-v1 in my submissions.
hendrycks/competition_math
Updated • 23.4k • 92Note This turned out to be one of the holdout tasks. mosaicml/instruct-v3 includes training problems from competition_math. This is a contributor to the high scores on GSM8k and MATH benchmarks.
kaist-ai/CoT-Collection
Viewer • Updated • 815 • 87Note Looked promising, but did not have time to explore.
tasksource/icl-symbol-tuning-instruct
Viewer • Updated • 4 • 16Note Considered for improving ICL. Did not have time to explore.
cais/mmlu
Viewer • Updated • 775k • 245Note Decided against training on MMLU data.
GAIR/lima
Viewer • Updated • 6.32k • 381Note Avoided due to CC BY-NC-SA license, though it would have been allowed for the competition. Likely would have been a good resource otherwise.
grammarly/coedit
Viewer • Updated • 3.72k • 45Note The plan here would be to target robustness metrics by finetuning an expert model to correct perturbations and/or clarify the input. This could have paraphrasing or other text revision tasks if they appeared in the hidden eval. Did not have time to fully explore.
leslyarun/c4_200m_gec_train100k_test25k
Viewer • Updated • 77 • 5Note Similar use case as coedit.
wanyu/IteraTeR_human_sent
Viewer • Updated • 5Note Similar use case as coedit.
allenai/social_i_qa
Updated • 191 • 7Note Now knowing that the holdout tasks had ethics questions, I wish I had used this.
lighteval/siqa
Viewer • Updated • 35.2k • 3Note Same as social_i_qa
tau/commonsense_qa
Viewer • Updated • 10k • 53Note Now knowing that the holdout tasks had ethics questions, I wish I had used this.
euirim/goodwiki
Viewer • Updated • 93 • 47Note Could have been useful for RAG.
alexfabbri/multi_news
Updated • 285 • 57Note The thought was this could help with CNN/DM summarization, but some quality and license concerns combined with acceptable performance without it led to its exclusion.
allenai/math_qa
Updated • 99 • 69allenai/ropes
Viewer • Updated • 55 • 42allenai/openbookqa
Viewer • Updated • 6.45k • 62allenai/ai2_arc
Viewer • Updated • 634k • 88INK-USC/riddle_sense
Updated • 300 • 21allenai/qasc
Viewer • Updated • 65 • 10nyu-mll/blimp
Viewer • Updated • 3.71k • 31google/boolq
Viewer • Updated • 5.91k • 55corypaik/prost
Viewer • Updated • 733 • 1allenai/sciq
Viewer • Updated • 3.56k • 77facebook/belebele
Viewer • Updated • 27.3k • 79derek-thomas/ScienceQA
Viewer • Updated • 6.74k • 111openlifescienceai/medmcqa
Viewer • Updated • 5.48k • 97embedding-data/QQP_triplets
Viewer • Updated • 509 • 4VMware/open-instruct
Viewer • Updated • 88 • 39