1 Training data← Back to the data viewer2 TokenizerCompareAuditExplainer slidesCoordinate tokens3 Image budget4 MP vs 8 MP4 NotesData processingAssamese ra checkSynthetic checkSmall held-out blocksData EDAJaigyan latticeJaigyan v4 re-readEmpty pagesManipuri: 4B vs pipeline5 TrainingEmbedding warm-upSFT ablationsSFT merges2B distillation0.8B distillationDistillation explainedRL planRL ablationsRL training

Manipuri pre-fill: compare versions

IndicDLP v4 pre-fill candidates, 3,870 Manipuri (Meitei Mayek) pages. Choose a version for each side: Layout + 4B = IndicDocLayout blocks (from bodhan-ai/indic-ocr) read by the 4B rl_best_ckpt with the block prompt; Pipeline = IndicDocLayout + 0.8B best_ckpt_rl per block; 4B page = 4B rl_best_ckpt, page prompt with headers, one pass. Hover a block to see its box. Boxes on a 0–1000 grid; dashed = child of a header/footer container.