CoolFace
Datasetpublic

Anushhh/KannadaPromptBench

KannadaPromptBench A benchmark dataset for evaluating prompt strategy sensitivity in Kannada, a low-resource Dravidian language. Dataset Summary Language: Kannada (kn) Tasks: Sentiment Analysis (100), Question Answering (75), Summarization (50) Total: 225 culturally grounded samples Inter-annotator agreement: Cohen's κ > 0.80 Dataset Structure Each sample contains: id, task, input_text, label, difficulty, domain. Citation Please… See the full description on the dataset page: https://huggingface.co/datasets/Anushhh/KannadaPromptBench.

sourceHugging Facecc-by-4.0updated 6mo agoView on Hugging Face
1likes33downloads
experiment_log.json256 linesDownload Raw Back to root
1{2  "experiment_name": "KannadaPromptBench-Enhanced",3  "version": "2.0",4  "date": "2026-04-12T09:58:03.861501",5  "models": [6    "llama3.1-8b",7    "sarvam-m",8    "mistralai/mistral-7b-instruct-v0.1"9  ],10  "strategies": [11    "zero_shot",12    "few_shot",13    "cot"14  ],15  "tasks": [16    "sentiment",17    "qa",18    "summarization"19  ],20  "total_dataset_size": 225,21  "total_results_rows": 2025,22  "session_csvs_used": [23    "cerebras_results.csv",24    "sarvam_results.csv",25    "openrouter_results.csv"26  ],27  "enhancements": {28    "inter_annotator_agreement": "Cohen kappa computed in Cell 4b",29    "statistical_significance": "McNemar test in Cell 9b",30    "error_analysis": "5-category analysis in Cell 11",31    "non_llm_baselines": "TF-IDF+LogReg in Cell 8a",32    "huggingface_export": "Cell 4c",33    "adversarial_robustness": "4 perturbation types in Cell 8b",34    "expanded_dataset": "100 sentiment + 75 QA + 50 summarization",35    "rouge_kannada_fix": "Character-level LCS fallback for Kannada ROUGE-L",36    "qa_exact_match_fix": "Substring match for verbose model responses"37  },38  "aggregate_results": [39    {40      "model": "llama3.1-8b",41      "strategy": "cot",42      "task": "qa",43      "accuracy": 0.747,44      "rouge_l": 0.189,45      "n_samples": 7546    },47    {48      "model": "llama3.1-8b",49      "strategy": "cot",50      "task": "sentiment",51      "accuracy": 0.32,52      "rouge_l": NaN,53      "n_samples": 10054    },55    {56      "model": "llama3.1-8b",57      "strategy": "cot",58      "task": "summarization",59      "accuracy": 0.12,60      "rouge_l": 0.402,61      "n_samples": 5062    },63    {64      "model": "llama3.1-8b",65      "strategy": "few_shot",66      "task": "qa",67      "accuracy": 0.747,68      "rouge_l": 0.679,69      "n_samples": 7570    },71    {72      "model": "llama3.1-8b",73      "strategy": "few_shot",74      "task": "sentiment",75      "accuracy": 0.55,76      "rouge_l": NaN,77      "n_samples": 10078    },79    {80      "model": "llama3.1-8b",81      "strategy": "few_shot",82      "task": "summarization",83      "accuracy": 0.64,84      "rouge_l": 0.61,85      "n_samples": 5086    },87    {88      "model": "llama3.1-8b",89      "strategy": "zero_shot",90      "task": "qa",91      "accuracy": 0.933,92      "rouge_l": 0.694,93      "n_samples": 7594    },95    {96      "model": "llama3.1-8b",97      "strategy": "zero_shot",98      "task": "sentiment",99      "accuracy": 0.5,100      "rouge_l": NaN,101      "n_samples": 100102    },103    {104      "model": "llama3.1-8b",105      "strategy": "zero_shot",106      "task": "summarization",107      "accuracy": 0.58,108      "rouge_l": 0.604,109      "n_samples": 50110    },111    {112      "model": "mistralai/mistral-7b-instruct-v0.1",113      "strategy": "cot",114      "task": "qa",115      "accuracy": 0.52,116      "rouge_l": 0.225,117      "n_samples": 75118    },119    {120      "model": "mistralai/mistral-7b-instruct-v0.1",121      "strategy": "cot",122      "task": "sentiment",123      "accuracy": 0.42,124      "rouge_l": NaN,125      "n_samples": 100126    },127    {128      "model": "mistralai/mistral-7b-instruct-v0.1",129      "strategy": "cot",130      "task": "summarization",131      "accuracy": 0.0,132      "rouge_l": 0.109,133      "n_samples": 50134    },135    {136      "model": "mistralai/mistral-7b-instruct-v0.1",137      "strategy": "few_shot",138      "task": "qa",139      "accuracy": 0.533,140      "rouge_l": 0.237,141      "n_samples": 75142    },143    {144      "model": "mistralai/mistral-7b-instruct-v0.1",145      "strategy": "few_shot",146      "task": "sentiment",147      "accuracy": 0.39,148      "rouge_l": NaN,149      "n_samples": 100150    },151    {152      "model": "mistralai/mistral-7b-instruct-v0.1",153      "strategy": "few_shot",154      "task": "summarization",155      "accuracy": 0.12,156      "rouge_l": 0.344,157      "n_samples": 50158    },159    {160      "model": "mistralai/mistral-7b-instruct-v0.1",161      "strategy": "zero_shot",162      "task": "qa",163      "accuracy": 0.587,164      "rouge_l": 0.185,165      "n_samples": 75166    },167    {168      "model": "mistralai/mistral-7b-instruct-v0.1",169      "strategy": "zero_shot",170      "task": "sentiment",171      "accuracy": 0.46,172      "rouge_l": NaN,173      "n_samples": 100174    },175    {176      "model": "mistralai/mistral-7b-instruct-v0.1",177      "strategy": "zero_shot",178      "task": "summarization",179      "accuracy": 0.12,180      "rouge_l": 0.394,181      "n_samples": 50182    },183    {184      "model": "sarvam-m",185      "strategy": "cot",186      "task": "qa",187      "accuracy": 0.6,188      "rouge_l": 0.439,189      "n_samples": 75190    },191    {192      "model": "sarvam-m",193      "strategy": "cot",194      "task": "sentiment",195      "accuracy": 0.34,196      "rouge_l": NaN,197      "n_samples": 100198    },199    {200      "model": "sarvam-m",201      "strategy": "cot",202      "task": "summarization",203      "accuracy": 0.02,204      "rouge_l": 0.121,205      "n_samples": 50206    },207    {208      "model": "sarvam-m",209      "strategy": "few_shot",210      "task": "qa",211      "accuracy": 0.707,212      "rouge_l": 0.137,213      "n_samples": 75214    },215    {216      "model": "sarvam-m",217      "strategy": "few_shot",218      "task": "sentiment",219      "accuracy": 0.37,220      "rouge_l": NaN,221      "n_samples": 100222    },223    {224      "model": "sarvam-m",225      "strategy": "few_shot",226      "task": "summarization",227      "accuracy": 0.0,228      "rouge_l": 0.061,229      "n_samples": 50230    },231    {232      "model": "sarvam-m",233      "strategy": "zero_shot",234      "task": "qa",235      "accuracy": 0.52,236      "rouge_l": 0.04,237      "n_samples": 75238    },239    {240      "model": "sarvam-m",241      "strategy": "zero_shot",242      "task": "sentiment",243      "accuracy": 0.4,244      "rouge_l": NaN,245      "n_samples": 100246    },247    {248      "model": "sarvam-m",249      "strategy": "zero_shot",250      "task": "summarization",251      "accuracy": 0.0,252      "rouge_l": 0.064,253      "n_samples": 50254    }255  ]256}