ldt-10m / README.md

Commit History

24 days ago

  1. Update to v4 checkpoint: 444M tokens, val 3.8943 (ppl 49.12), still incoherent

    Verified signature
    ba470c9
  2. Fix fabricated sample quote + overstatement: replace sample #1 (not in eval output) with real samples from eval_ldt10m_final.json; correct "coherent/fluency" to "grammatically structured but semantically incoherent" (#10)⁠

    451f582
  3. Add LDT-10M model card (#7)⁠

    1a279d1
  4. Add honest model card (architecture, training, undertraining status, real samples)

    Verified signature
    fc4ad9c