{
 "body": "\nPart of the [FewCLUE](fewclue.md) family.\n\n## What it measures\n\nOCNLI-FC is FewCLUE's few-shot cut of OCNLI, the CLUE benchmark's Original Chinese Natural Language\nInference task: given a premise and a hypothesis sentence, the model classifies their relationship as\nentailment, neutral, or contradiction. OCNLI's premises are drawn from five genres (news, government\ndocuments, fiction, TV transcripts and telephone transcripts) and its hypotheses were written by\nhired Chinese-major university students following MNLI's methodology, making it a natively authored\nChinese NLI set rather than a translation. The \"-FC\" name, used in OpenCompass's harness config and\nin this wiki, distinguishes this 32-example few-shot slice from CLUE's full-size OCNLI training data.\n\n## Reading the numbers\n\nThe labelled public test set holds 2,520 pairs, scored by accuracy against a 38.1% three-way majority\nbaseline (above the 33.3% uniform rate, reflecting class imbalance). Human evaluators scored 90.3%;\nthe strongest few-shot method reported is EFL at 66.2%, well ahead of the 41-44% that cloze-based\nmethods (PET, P-tuning) managed here, even though those led on most other FewCLUE tasks --\nentailment-style fine-tuning transfers naturally to an inference task. A score in the low 40s\nsuggests a cloze-prompted method; one near or above the mid-60s suggests an entailment-trained one.\nSee the [FewCLUE](fewclue.md) family page for the shared protocol and contamination notes that apply\nhere too.\n",
 "build": {
  "built_at": "2026-09-09T16:56:50+00:00",
  "commit": "0a599558854c0e238c03a0f0d725239cb28f9d11",
  "eligibility_as_of": "2026-09-09"
 },
 "disposition": {
  "canonical_id": "fewclue_ocnli_fc",
  "reasons": [],
  "status": "unassessed",
  "verified_results": []
 },
 "models_covered": [],
 "page": {
  "category": "reasoning",
  "dataset": {
   "languages": [
    "zh"
   ],
   "license": "",
   "modalities": [
    "text"
   ],
   "public_test_set": true,
   "size": 2520,
   "size_note": "2,520 labelled public test pairs (test_public.json, used for scoring), plus 3,000 in the private test set (original leaderboard only), 32 train and 32 dev examples per split (5 parallel splits, train_0..train_4 / dev_0..dev_4, plus a merged train_few_all/dev_few_all), and 20,000 unlabelled pairs not used for scoring. Figures from the FewCLUE paper's Table 1, matching the GitHub README. Underlying data is OCNLI (Hu et al. 2020), the CLUE benchmark's natively-authored Chinese NLI task; OpenCompass's harness config loads this task's data from the same \"ocnli\" directory as full-size OCNLI, confirming the two share source data at different training-set sizes.\n",
   "splits": "train_0..train_4 (32 each) + train_few_all; dev_0..dev_4 (32 each) + dev_few_all; test_public (2,520, labelled); test (3,000, private); unlabeled (20,000)",
   "url": "https://github.com/CLUEbenchmark/FewCLUE/tree/main/datasets/ocnli"
  },
  "freshness": {
   "researched": "2026-09-08",
   "researched_by": "sonnet-5 agent, batch 5, slice B"
  },
  "harness": {
   "opencompass": "FewCLUE_ocnli_fc"
  },
  "id": "fewclue_ocnli_fc",
  "leaderboard_url": "https://www.cluebenchmarks.com/fewclue.html",
  "lineage": {
   "family": "fewclue"
  },
  "measures": "A premise and hypothesis sentence pair, natively authored in Chinese across five genres; the model classifies their relationship as entailment, neutral or contradiction, learned few-shot from 32 labelled training pairs -- the few-shot cut of the CLUE benchmark's OCNLI task.\n",
  "metric": {
   "baseline_note": "38.1% is the paper's majority-class baseline, somewhat above the three-way uniform-random rate of 33.3%, reflecting a class imbalance in the test set. See the fewclue family page for the top overall few-shot method scores.\n",
   "direction": "higher_is_better",
   "human_baseline": 90.3,
   "max_score": 100,
   "name": "accuracy",
   "random_baseline": 38.1,
   "unit": "%"
  },
  "name": "FewCLUE: OCNLI-FC (Natural Language Inference)",
  "page_kind": "subset",
  "paper": {
   "arxiv": "2107.07498",
   "title": "FewCLUE: A Chinese Few-shot Learning Evaluation Benchmark",
   "url": "https://arxiv.org/abs/2107.07498",
   "year": 2021
  },
  "publisher": {
   "authors": [
    "Liang Xu",
    "Xiaojing Lu",
    "Chenyang Yuan",
    "Xuanwei Zhang",
    "Huilin Xu",
    "Hu Yuan",
    "Guoao Wei",
    "Xiang Pan",
    "Xin Tian",
    "Libo Qin",
    "Hu Hai"
   ],
   "org": "CLUE team",
   "url": "https://github.com/CLUEbenchmark/FewCLUE"
  },
  "released": "2021-04",
  "repo_url": "https://github.com/CLUEbenchmark/FewCLUE",
  "sources": [
   {
    "accessed": "2026-09-08",
    "title": "FewCLUE: A Chinese Few-shot Learning Evaluation Benchmark (Xu et al., arXiv:2107.07498)",
    "url": "https://arxiv.org/abs/2107.07498"
   },
   {
    "accessed": "2026-09-08",
    "title": "CLUEbenchmark/FewCLUE GitHub repository (task description, dataset statistics)",
    "url": "https://github.com/CLUEbenchmark/FewCLUE"
   },
   {
    "accessed": "2026-09-08",
    "title": "OpenCompass FewCLUE_ocnli_fc dataset config",
    "url": "https://raw.githubusercontent.com/open-compass/opencompass/main/opencompass/configs/datasets/FewCLUE_ocnli_fc/FewCLUE_ocnli_fc_gen_f97a97.py"
   }
  ],
  "status": "unknown",
  "subcategory": "three-way natural language inference over native Chinese sentence pairs, few-shot cut of OCNLI",
  "summary": "FewCLUE's few-shot cut of OCNLI: classify a Chinese premise-hypothesis pair as entailment, neutral or contradiction, learned from 32 labelled training pairs.",
  "tags": [
   "chinese",
   "few-shot",
   "fewclue-subset",
   "nli"
  ],
  "task_format": "Three-way natural language inference (entailment / neutral / contradiction), graded on the single correct label; evaluated from a 32-example few-shot training split, one of five parallel splits FewCLUE provides for this task.\n"
 }
}