comprehensive_results.json

{
  "experiment_info": {
    "name": "Climate Education Chatbot - Optimal Model",
    "version": "Experiment 4c - Balanced",
    "description": "Best performing model across all experiments",
    "timestamp": "2025-06-17T00:58:29.482424"
  },
  "model_performance": {
    "training_loss": 0.5757,
    "validation_loss": 0.8844,
    "bleu_score": 0.0549,
    "average_generation_time": 17.3,
    "performance_grade": "OPTIMAL"
  },
  "experiment_progression": {
    "baseline": {
      "learning_rate": 2e-05,
      "batch_size": 16,
      "epochs": 5,
      "train_loss": 9.04,
      "val_loss": 9.79,
      "bleu": 0.0001,
      "notes": "Severe underfitting"
    },
    "experiment_2": {
      "learning_rate": 0.0005,
      "batch_size": 8,
      "epochs": 10,
      "train_loss": 0.93,
      "val_loss": 0.93,
      "bleu": 0.0012,
      "notes": "Major leap in text quality"
    },
    "experiment_3": {
      "learning_rate": 0.0003,
      "batch_size": 4,
      "epochs": 15,
      "train_loss": 1.2623,
      "val_loss": 1.3637,
      "bleu": 0.0392,
      "notes": "Smooth fluency; minor factual errors"
    },
    "experiment_4": {
      "learning_rate": 0.0001,
      "batch_size": 8,
      "epochs": 20,
      "train_loss": 0.5757,
      "val_loss": 0.8844,
      "bleu": 0.0392,
      "notes": "Best factual grounding, improved conciseness"
    },
    "experiment_4b": {
      "description": "Speed-up variant",
      "bleu": 0.034,
      "generation_time": 16.7,
      "notes": "Speed-up variant; less repetition but BLEU dropped"
    },
    "experiment_4c": {
      "description": "OPTIMAL - Balanced version",
      "bleu": 0.0549,
      "generation_time": 17.3,
      "notes": "BEST: Stable quality + fast generation"
    }
  },
  "optimal_parameters": {
    "max_length": 70,
    "min_length": 18,
    "temperature": 0.5,
    "top_p": 0.8,
    "top_k": 40,
    "repetition_penalty": 2.0,
    "no_repeat_ngram_size": 3,
    "num_beams": 1
  },
  "dataset_info": {
    "training_samples": 60,
    "validation_samples": 13,
    "test_samples": 13,
    "domain": "Climate Education",
    "model_type": "T5-small generative QA"
  },
  "achievements": [
    "93% reduction in training loss (9.04 \u2192 0.5757)",
    "55x improvement in BLEU score (0.0001 \u2192 0.0549)",
    "30% speed improvement from baseline",
    "Balanced quality and performance",
    "Functional interactive chatbot"
  ]
}