pythia-70m-deduped-cleansharegpt-en/README.md

license, datasets, language

license	datasets	language
apache-2.0	shibing624/sharegpt_gpt4	en

Pythia-70m-deduped finetuned on a cleaned version of ShareGPT data. The cleaned dataset is obtained by removing duplicates and paraphrases from the original corpus, and keeping only the English instance. The final training size is of 3507 instances.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	25.06
ARC (25-shot)	21.16
HellaSwag (10-shot)	27.16
MMLU (5-shot)	25.24
TruthfulQA (0-shot)	48.57
Winogrande (5-shot)	50.12
GSM8K (5-shot)	0.0
DROP (3-shot)	3.15