Practice insights

Which hyperparameter values are common vs rare — and which correlate with stronger HuggingFace outcomes. Scores are population percentiles in 0–1; the works/fails axis is stratified by base model.

Models

HuggingFace LoRA models with outcome signals and the composite success score. Click a row for its extracted training recipe.

modelbasedownloads likesfan-outscore

Repositories

GitHub repos with extracted LoRA configs. Click a row for the recipe and any linked HuggingFace models.

repositorylanguagestars paramsconfidence