experiment_log.json

{
  "experiment_name": "KannadaPromptBench-Enhanced",
  "version": "2.0",
  "date": "2026-04-12T09:58:03.861501",
  "models": [
    "llama3.1-8b",
    "sarvam-m",
    "mistralai/mistral-7b-instruct-v0.1"
  ],
  "strategies": [
    "zero_shot",
    "few_shot",
    "cot"
  ],
  "tasks": [
    "sentiment",
    "qa",
    "summarization"
  ],
  "total_dataset_size": 225,
  "total_results_rows": 2025,
  "session_csvs_used": [
    "cerebras_results.csv",
    "sarvam_results.csv",
    "openrouter_results.csv"
  ],
  "enhancements": {
    "inter_annotator_agreement": "Cohen kappa computed in Cell 4b",
    "statistical_significance": "McNemar test in Cell 9b",
    "error_analysis": "5-category analysis in Cell 11",
    "non_llm_baselines": "TF-IDF+LogReg in Cell 8a",
    "huggingface_export": "Cell 4c",
    "adversarial_robustness": "4 perturbation types in Cell 8b",
    "expanded_dataset": "100 sentiment + 75 QA + 50 summarization",
    "rouge_kannada_fix": "Character-level LCS fallback for Kannada ROUGE-L",
    "qa_exact_match_fix": "Substring match for verbose model responses"
  },
  "aggregate_results": [
    {
      "model": "llama3.1-8b",
      "strategy": "cot",
      "task": "qa",
      "accuracy": 0.747,
      "rouge_l": 0.189,
      "n_samples": 75
    },
    {
      "model": "llama3.1-8b",
      "strategy": "cot",
      "task": "sentiment",
      "accuracy": 0.32,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "llama3.1-8b",
      "strategy": "cot",
      "task": "summarization",
      "accuracy": 0.12,
      "rouge_l": 0.402,
      "n_samples": 50
    },
    {
      "model": "llama3.1-8b",
      "strategy": "few_shot",
      "task": "qa",
      "accuracy": 0.747,
      "rouge_l": 0.679,
      "n_samples": 75
    },
    {
      "model": "llama3.1-8b",
      "strategy": "few_shot",
      "task": "sentiment",
      "accuracy": 0.55,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "llama3.1-8b",
      "strategy": "few_shot",
      "task": "summarization",
      "accuracy": 0.64,
      "rouge_l": 0.61,
      "n_samples": 50
    },
    {
      "model": "llama3.1-8b",
      "strategy": "zero_shot",
      "task": "qa",
      "accuracy": 0.933,
      "rouge_l": 0.694,
      "n_samples": 75
    },
    {
      "model": "llama3.1-8b",
      "strategy": "zero_shot",
      "task": "sentiment",
      "accuracy": 0.5,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "llama3.1-8b",
      "strategy": "zero_shot",
      "task": "summarization",
      "accuracy": 0.58,
      "rouge_l": 0.604,
      "n_samples": 50
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "cot",
      "task": "qa",
      "accuracy": 0.52,
      "rouge_l": 0.225,
      "n_samples": 75
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "cot",
      "task": "sentiment",
      "accuracy": 0.42,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "cot",
      "task": "summarization",
      "accuracy": 0.0,
      "rouge_l": 0.109,
      "n_samples": 50
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "few_shot",
      "task": "qa",
      "accuracy": 0.533,
      "rouge_l": 0.237,
      "n_samples": 75
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "few_shot",
      "task": "sentiment",
      "accuracy": 0.39,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "few_shot",
      "task": "summarization",
      "accuracy": 0.12,
      "rouge_l": 0.344,
      "n_samples": 50
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "zero_shot",
      "task": "qa",
      "accuracy": 0.587,
      "rouge_l": 0.185,
      "n_samples": 75
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "zero_shot",
      "task": "sentiment",
      "accuracy": 0.46,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "mistralai/mistral-7b-instruct-v0.1",
      "strategy": "zero_shot",
      "task": "summarization",
      "accuracy": 0.12,
      "rouge_l": 0.394,
      "n_samples": 50
    },
    {
      "model": "sarvam-m",
      "strategy": "cot",
      "task": "qa",
      "accuracy": 0.6,
      "rouge_l": 0.439,
      "n_samples": 75
    },
    {
      "model": "sarvam-m",
      "strategy": "cot",
      "task": "sentiment",
      "accuracy": 0.34,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "sarvam-m",
      "strategy": "cot",
      "task": "summarization",
      "accuracy": 0.02,
      "rouge_l": 0.121,
      "n_samples": 50
    },
    {
      "model": "sarvam-m",
      "strategy": "few_shot",
      "task": "qa",
      "accuracy": 0.707,
      "rouge_l": 0.137,
      "n_samples": 75
    },
    {
      "model": "sarvam-m",
      "strategy": "few_shot",
      "task": "sentiment",
      "accuracy": 0.37,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "sarvam-m",
      "strategy": "few_shot",
      "task": "summarization",
      "accuracy": 0.0,
      "rouge_l": 0.061,
      "n_samples": 50
    },
    {
      "model": "sarvam-m",
      "strategy": "zero_shot",
      "task": "qa",
      "accuracy": 0.52,
      "rouge_l": 0.04,
      "n_samples": 75
    },
    {
      "model": "sarvam-m",
      "strategy": "zero_shot",
      "task": "sentiment",
      "accuracy": 0.4,
      "rouge_l": NaN,
      "n_samples": 100
    },
    {
      "model": "sarvam-m",
      "strategy": "zero_shot",
      "task": "summarization",
      "accuracy": 0.0,
      "rouge_l": 0.064,
      "n_samples": 50
    }
  ]
}