1
Training data
← Back to the data viewer
2
Tokenizer
Compare
Audit
Explainer slides
Coordinate tokens
3
Image budget
4 MP vs 8 MP
4
Notes
Data processing
Assamese ra check
Synthetic check
Small held-out blocks
Data EDA
Jaigyan lattice
Jaigyan v4 re-read
Empty pages
Manipuri: 4B vs pipeline
5
Training
Embedding warm-up
SFT ablations
SFT merges
2B distillation
0.8B distillation
Distillation explained
RL plan
RL ablations
RL training
Training data EDA (train split)
Metric
Rows
per set
by domain
by type (pr/hw/…)
by kind
by language
by origin
by section
Section
all
unified-pages
block-curriculum
Click up to 4 rows to compare their distributions (share of rows per log-spaced bin). Unified sets: exact values for every train page; block sets: a random sample per set.