Release fixed MetaScore-SFT-100 benchmark and source assets with per-piece attribution
42d9e00 verified Download reconstruction/preprocess/config.py from naivecynics/MetaScore-SFT-100: direct link, hf CLI and curl.
- Browser
- Download file 519 Bytes
-
https://huggingface.co/datasets/naivecynics/MetaScore-SFT-100/resolve/main/reconstruction/preprocess/config.py
- Command line
-
hf download hf://datasets/naivecynics/MetaScore-SFT-100/reconstruction/preprocess/config.py
-
curl -L -o config.py https://huggingface.co/datasets/naivecynics/MetaScore-SFT-100/resolve/main/reconstruction/preprocess/config.py
519 Bytes
| """Portable settings for the archived fixed-window builder. | |
| The window serialization code is unchanged. Only deployment-specific pipeline | |
| paths are replaced with relative placeholders; build_metascore_test uses its | |
| explicit --source and --output arguments. | |
| """ | |
| from pathlib import Path | |
| DATASETS = ('all',) | |
| RAW_DIR = Path('raw') | |
| INDEX_DIR = Path('index') | |
| SLICES_DIR = Path('slices') | |
| PIPELINE_VERSION = 4 | |
| REQUIRED_CATEGORIES = ('single_piano', 'contains_piano') | |
| REQUIRE_4_4 = False | |
| SOURCES_PER_SHARD = 1000 | |
| WORKERS = 1 | |