Raw generated text samples used in the ranking_divergence
rank-histogram divergence metric exploration (andreamiele/ranking_divergence, branch Andrea-dev),
sampled from DUO, MDLM, and S-FLM (SFM) using the official s-flm repo/checkpoints.
duo-sflm-v1/ — 245 configs, temperature 0.2-1.4 x NFE {8,16,32,64,128}
mdlm-sflm-v1/ — 155 configs, temperature 0.1-2.0 x NFE {8,16,32,64,128}
sfm-sflm-duotemps-v1/ — 245 configs, temperature… See the full description on the dataset page:
https://huggingface.co/datasets/aerdna/ranking-divergence-samples.
