v51: composite=88.68 — see model card for benchmark deltas vs v45

v45: SFT+chained GRPO with ITN — 95.9% number accuracy, 97.0% filler-free, deletion behavior matches v36

v36: full-FT GRPO with substantive-deletion-aware reward — filler-free 96.9%, sub-del-15-long 0.64%

Update README.md

v23 R6 (4-stage SFT->GRPO->Stage2->GRPO): ROUGE-L 0.9537 (tied v22), Filler-Free 91.1% (beats v22 90.3%), paragraph rate 91.5% — definitive v23 model

v23 R5 (paragraph rows excluded from GRPO): ROUGE-L 0.9505, Filler-Free 91.0%, paragraph rate 89.5% — best v23 variant overall

v23 R4 (LR 5e-6): Filler-Free 91.0% (beats v22 90.3%), paragraph rate 89%, ROUGE-L 0.9499

v23+paragraphs: ROUGE-L 0.9506, Filler-Free 90.2%, paragraph rate 91.5% (0% in v22)

v22+GRPO-r32: 5-bit MLX

v22+GRPO: 5-bit MLX (0.953 val)

v22-lr3: 5-bit MLX (0.948 val set)

v18: 5-bit MLX (0.968 parent)

v17: 5-bit MLX (0.966 parent)

v16: 5-bit MLX quant (ROUGE-L 0.962 parent)

fix: use Unicode star emoji instead of :star: shortcode

Update model card: v15 detailed card with examples, benchmarks, and links

v15: 5-bit MLX quant (237MB, ROUGE-L ~0.955)

v7: 5-bit MLX, ROUGE-L 0.943

v5: ROUGE-L 0.942, 5-bit MLX 233MB

v4: 5-bit MLX, 233MB, long transcript support

Update README.md

Upload README.md with huggingface_hub

5-bit MLX: ROUGE-L 0.926, 233MB, 56% exact, 99% filler-free

initial commit