hamishivi/tulu-2-multitask-rrmax-326k-sft

Updated Mar 4 • 1

Note Above is our Llama 2 7b model trained on the multitask mixture linked below on Tulu 2 data.

hamishivi/llama-3.1-tulu-3-multitask-rrmax-939k-sft

Updated Mar 4 • 2

Note Above is our Llama 3 8b model trained on the multitask mixture linked below on Tulu 3 data.

Note Below is our unfiltered datasets, multi-million size instruction tuning datasets made up of all the data considered for Tulu 2 and 3 respectively.

Note Below are other multitask trained models and the data they were trained on. Tulu 2 models are based on Llama 2 7b and Tulu 3 models on Llama 3 8b.

Note Below is the data selected by RDS+ with Llama 2 7b from the Tulu 2 unfiltered dataset, selecting for the evaluation in the dataset name.