benchmark v2 — E2E 장면 100장 (15/80/5, 막대 비율 2.0-2.3 / 5:3·6:4 반반, 포장 ±15°, seed 2027, code cc9e87a) (part 2)

benchmark v2 — E2E 장면 100장 (15/80/5, 막대 비율 2.0-2.3 / 5:3·6:4 반반, 포장 ±15°, seed 2027, code cc9e87a)

benchmark v1 — E2E 장면 100장 (first_ok 15 / target 80 / hard 5, seed 2026, code fe4ddb0) (part 2)

benchmark v1 — E2E 장면 100장 (first_ok 15 / target 80 / hard 5, seed 2026, code fe4ddb0)

카드 정정 — 학습 JSONL 은 렌더 전량 기록이다 (로더가 band 로 고른다)

Upload folder using huggingface_hub (part 4)

Upload folder using huggingface_hub (part 3)

Upload folder using huggingface_hub (part 2)

Upload folder using huggingface_hub

Upload folder using huggingface_hub (part 4)

Upload folder using huggingface_hub (part 3)

Upload folder using huggingface_hub (part 2)

Upload folder using huggingface_hub

Upload folder using huggingface_hub (part 4)

Upload folder using huggingface_hub (part 3)

Upload folder using huggingface_hub (part 2)

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload synthetic/v1/recipes/train.H.jsonl.gz with huggingface_hub

Upload synthetic/v1/recipes/train.M.jsonl.gz with huggingface_hub

Upload synthetic/v1/recipes/train.L.jsonl.gz with huggingface_hub

Upload README.md with huggingface_hub

initial commit