Download dataset/dataset_meta.json from tencent/GroupAssistBench: direct link, hf CLI and curl.
- Browser
- Download file 2.24 kB
-
https://huggingface.co/datasets/tencent/GroupAssistBench/resolve/main/dataset/dataset_meta.json
- Command line
-
hf download hf://datasets/tencent/GroupAssistBench/dataset/dataset_meta.json
-
curl -L -o dataset_meta.json https://huggingface.co/datasets/tencent/GroupAssistBench/resolve/main/dataset/dataset_meta.json
2.24 kB
| { | |
| "name": "GroupAssistBench final evaluation set", | |
| "built_at": "2026-11-03 20:13:25", | |
| "n_items": 2395, | |
| "n_groups": 50, | |
| "layer": { | |
| "L1": 1374, | |
| "L2": 1021 | |
| }, | |
| "difficulty": { | |
| "D1": 997, | |
| "D2": 1098, | |
| "D3": 300 | |
| }, | |
| "cell": { | |
| "L1D1": 776, | |
| "L1D2": 448, | |
| "L2D1": 221, | |
| "L2D2": 650, | |
| "L1D3": 150, | |
| "L2D3": 150 | |
| }, | |
| "query_origin": { | |
| "native": 1159, | |
| "authored": 1236 | |
| }, | |
| "track": { | |
| "jsonl": 864, | |
| "xlsx": 1531 | |
| }, | |
| "d3v2_items": 153, | |
| "files": { | |
| "items.jsonl": "evaluation records (query, gt / gt_atoms / distractor_atoms for L1, criteria for L2, cut_idx, mem_span)", | |
| "items_full.jsonl": "construction records of the same 2395 items: anchors (turn, fact, type, evidence), provenance, pos_turn/src_turn, event_id, pairing", | |
| "contexts.json": "{group: [[turn, speaker, text], ...]} full anonymised conversation, text cut at 2000 chars", | |
| "cutoffs.json": "{iid: cut_idx} a system may read only turns before this one (turn < cut_idx)", | |
| "profiles.json": "{group: [{member, slot, fact, turn}, ...]} member profiles; entries with turn > 0 are extracted from the conversation and point to the turn that establishes them, entries with turn = 0 are member background information", | |
| "manifest.json": "{\"<group>/<L>/<D>\": [iid, ...]} the 2395 ids bucketed by group, layer and difficulty" | |
| }, | |
| "provenance": { | |
| "base_items": { | |
| "n": 2242, | |
| "note": "selected from the step-04 pool of the 50 group runs (section 7 of the README)" | |
| }, | |
| "d3v2_items": { | |
| "n": 153, | |
| "note": "cut_idx may be moved earlier than the construction record (fallback), never later" | |
| } | |
| }, | |
| "sha256": { | |
| "items.jsonl": "3bfea2e0ec259b0eaf7681ffb622b736aff17b62e3422e8c65ffd17dce0f3ad8", | |
| "items_full.jsonl": "6364f1b3ec16c4b31b36e222bd50e618b23051f876a08007b4603abc13602a8a", | |
| "contexts.json": "0f075be152eab1dbe0123c6b76a81c003b97ec1e99e804c7cb85753b576a3638", | |
| "cutoffs.json": "467ee0cea8e3c486e93b1727f696cfde36307ca100ee40f8d1c5b52f4db3cb9e", | |
| "profiles.json": "84e8fc107f602a807b23887181c07371d71c2ad3fc145e3bab30af850dbbc847", | |
| "manifest.json": "d040af5f0fccad229b0ea85d50298d5875f80df2a9353cb7beb27809bc0ee19e" | |
| }, | |
| "annotator_ids": "human labels are recorded as anonymous ids (human:annotator_a / human:annotator_b)" | |
| } | |