Add sampled_id_list.json (33-definition dtype-symmetric subset) and document it in the dataset card
#3
by VanishD - opened
sampled_id_list.json names a 33-definition representative subset of the 103 definitions in definitions/ (ncnn 12/30, llama.cpp 8/26, simd-loop 13/47).
It is dtype-symmetric: each sampled op type appears in every dtype this dataset provides for it, on identical shapes, so a per-dtype comparison within one op type is not confounded by which shapes were sampled.
The dataset card gains a 'Sampled subset' section and a line in the repository-structure block.
VanishD changed pull request status to closed