AstrAI-paper

T

ViperEkura e6be33aa53 doc: add SFT length-filter rationale — high PPL + high variance from A.3 IFD figure

- Add 15-token length floor to SFT samples, with explicit reference
  to Appendix A.3 / Figure 5 (ifd_length_grid)
- Short replies (<10 tokens) show both high per-token perplexity
  (L_uncond ~6-8, PPL ~400-3000 vs long replies ~2-3, PPL ~7-20)
  and wide variance in L_cond (span 0-17.5), which distorts
  downstream IFD-based difficulty estimates.

2026-07-24 06:37:35 +08:00

data

fix: align SFT/DPO figures and text with actual training data

2026-07-24 06:15:42 +08:00

.gitignore

Restructure data pipeline: replace IFD analysis with length bias appendix

2026-07-05 18:59:29 +08:00

main.tex

doc: add SFT length-filter rationale — high PPL + high variance from A.3 IFD figure

2026-07-24 06:37:35 +08:00