evaluation/results/final_perplexity.json

{
  "evaluated_at": "2026-04-30T12:36:28.570037",
  "test_data": "data/processed/test.jsonl",
  "seq_len": 512,
  "results": [
    {
      "model": "Qwen2.5-0.5B",
      "perplexity": 5.3847,
      "mean_nll": 1.683567,
      "samples": 500,
      "status": "success"
    },
    {
      "model": "Hindi GPT-2 (ours)",
      "perplexity": 29.2873,
      "mean_nll": 3.377154,
      "tokens_evaluated": 4830208,
      "checkpoint_step": 12000,
      "status": "success"
    }
  ]
}