Datasets:
Download src/explicit_learning/ingest/mmr1.py from sungguk/visual-answerability: direct link, hf CLI and curl.
- Browser
- Download file 2.56 kB
-
https://huggingface.co/datasets/sungguk/visual-answerability/resolve/main/src/explicit_learning/ingest/mmr1.py
- Command line
-
hf download hf://datasets/sungguk/visual-answerability/src/explicit_learning/ingest/mmr1.py
-
curl -L -o mmr1.py https://huggingface.co/datasets/sungguk/visual-answerability/resolve/main/src/explicit_learning/ingest/mmr1.py
2.56 kB
| """MMR1-RL importer — composite multi-image rows with inferred provenance. | |
| MMR1-RL aggregates several public benchmarks but does not expose trustworthy | |
| row-level source IDs (``docs/01`` §3.2). Any source hint found on a row is | |
| recorded as ``provenance.inferred_source`` and is never treated as a definitive | |
| ID. Rows may carry multiple images; their order is preserved exactly and each | |
| image is content-addressed separately (``docs/02`` §2). | |
| Decontamination is required for every MMR1 row; that is enforced in P3, but the | |
| importer records the composite approval gate so the policy is visible downstream. | |
| """ | |
| from __future__ import annotations | |
| from collections.abc import Mapping | |
| from typing import Any | |
| from .base import ( | |
| ImageResolver, | |
| ImageStore, | |
| IngestError, | |
| NormalizedItem, | |
| Policy, | |
| infer_open_answer_type, | |
| make_item, | |
| ) | |
| NAME = "mmr1_rl" | |
| POLICY: Policy = "c2_train_candidate" | |
| LICENSE_GATE = "mmr1_composite_terms" | |
| ALLOWED_SPLIT = "train" | |
| def normalize( | |
| row: Mapping[str, Any], | |
| images: ImageStore, | |
| resolve: ImageResolver, | |
| *, | |
| revision: str, | |
| split: str, | |
| config: str = "default", | |
| ) -> NormalizedItem: | |
| """Normalize one MMR1-RL native row to a :class:`NormalizedItem`.""" | |
| if split != ALLOWED_SPLIT: | |
| raise IngestError(f"mmr1_rl: only split {ALLOWED_SPLIT!r} may be ingested, got {split!r}") | |
| question = str(row["problem"]) | |
| answer_raw = str(row["answer"]) | |
| answer_type = infer_open_answer_type(answer_raw) | |
| image_refs = [str(r) for r in row["images"]] | |
| if not image_refs: | |
| raise IngestError(f"mmr1_rl: row {row.get('id', '?')!r} has no images") | |
| image_paths: list[str] = [] | |
| image_sha: list[str] = [] | |
| for ref in image_refs: | |
| rel, digest = images.store(resolve(ref)) | |
| image_paths.append(rel) | |
| image_sha.append(digest) | |
| inferred = row.get("source") or row.get("inferred_source") | |
| extra = {"inferred_source": str(inferred) if inferred is not None else None} | |
| native_id = str(row.get("id", row.get("index"))) | |
| return make_item( | |
| source=NAME, | |
| source_revision=revision, | |
| source_config=config, | |
| source_split=split, | |
| source_native_id=native_id, | |
| question=question, | |
| choices=(), | |
| answer_raw=answer_raw, | |
| answer_canonical=answer_raw, | |
| answer_type=answer_type, | |
| image_paths=tuple(image_paths), | |
| image_sha256=tuple(image_sha), | |
| policy=POLICY, | |
| native_row=row, | |
| license_gate=LICENSE_GATE, | |
| extra_provenance=extra, | |
| ) | |