Note # of samples = 187,005 | 180,462 in paper

Note # of samples = 33,955 | 30,106 in paper

Note # of samples = 40,625 | 64,246 in paper

Note # of samples = 11,450 | 11,290 in paper

Note # of samples = 821,007 | 17,721 in paper I took it from medalpaca, not allenai.

Note # of samples = 272,458 | 499 in paper Maybe authors used dev dataset, which consists of 500 data

Note # of samples = 2,657 | 2,657 in paper

Note # of samples = 2,208 | 1,950 in paper

Note # of samples = 11,679 | 11,679 in paper

Note # of samples = 2,446 | 2,169 in paper

Note # of samples = 31,532 | 66,026 in paper

Note # of samples = 16,407 | 14,553 in paper

Note # of samples = 435 | 244 in paper

Note From https://github.com/allenai/natural-instructions, I filtered out json files with Domains "Healthcare", "Medicine", and "Scientific Research Papers". Note that this is not a data to be used for actual training, but a kind of metadata. # of samples 11,447 in paper

Note # of samples = 9,804 | 9,804 in paper

Note # of samples = 106,555 | 120,000 in paper ACI-Bench link: https://github.com/wyim/aci-bench # of samples = 87 | 87 in paper

Note # of samples = 1,301 | 2,602 in paper [General Domain] I'm not sure what is SlimOrca T0 and SlimOrca CoT. Maybe authors sampled from OpenOrca, not SlimOrca.

Note # of samples = 292,576 in paper

Note # of samples = 50,953 in paper [DPO dataset]