sgaofen / humanizer-local-model

What it does

Four drafts from the held-out evaluation set (never seen in training). For each draft we generated 8 samples with this release's humanizer-12b-Q8_0.gguf file (llama.cpp, the app's settings) and picked the one that reads best among those the fact judge passed. The right side is that sample, not edited; only whitespace is normalised for display. All 8 samples per draft are in eval/outputs/examples-12b-Q8_0_x8.json. Highlight = new wording, strikethrough = draft wording that was replaced. We also checked every number and name in these four by hand.