iamPi/orrery-ac54af81
halyard-ac54af81 Format-repaired SFT corpus for Bittensor SN120 (Affine), Reason v4 (weight_version_key=7). TRL prompt-completion format. What was repaired Every row's thought channel was normalised to exactly ` THOUGHT: {{z}} {{y}}`. 40.08% of rows (7,270 / 18,138) carried at least one defect: defect rows stray <think> / </think> / <thinking> / </thinking> inside the thought 7,124 corrupted THOUGHT label (THO, THOTH:, THOFTWARE:, THOTHought:, THOURLY… See the full description on the dataset page: https://huggingface.co/datasets/iamPi/orrery-ac54af81.
halyard-ac54af81
Format-repaired SFT corpus for Bittensor SN120 (Affine), Reason v4 (weight_version_key=7). TRL prompt-completion format.
What was repaired
Every row's thought channel was normalised to exactly `</think> THOUGHT: {{z}}
{{y}}`. 40.08% of rows (7,270 / 18,138) carried at least one defect:
Nothing was dropped. Tags and labels are markers, so removing them leaves the reasoning intact. 61 rows carry up to 8,077 characters of real analysis before their second label, so a "keep only what follows the last THOUGHT:" rule would have deleted it.
Row set, prompts and actions are byte-identical to `iamPi/tessera-77d11909`; only the thought text changed. Every row was re-parsed with the validator's own split_rollout() and checked byte-for-byte against the force_text() rendering before writing.
Known caveats
- 444 row(s) fall below the 80-character crown floor. Their length was previously made up partly by the junk that has now been removed.
- 0 row(s) trip the validator's fuzzy leakage heuristic.
- 28 row(s) exceed the 1,792-token rollout budget — pre-existing in tessera, not introduced here.
Built with scripts/normalize_sft_format.py.