{
  "command": "train.py --corpus-dir data/glotlid_group_21 --initial-vocab /workspace/hf_home/hub/models--CohereLabs--tiny-aya-global/snapshots/b2f40192d990d66a6f28a4b90d72817ab9c613da/tokenizer.json --per-lang-counts-method soft --em-iterations 10 --results-dir results_tinyaya_soft_full --lang-batch-size 5 --max-lang-samples-per-lang 100000 --reuse-base",
  "timestamp": "2026-03-23T14:30:59.285253+00:00",
  "training_completed": true,
  "source": {
    "format": "corpus",
    "path": "/workspace/runpod/tinyAyaLid/data/glotlid_group_21",
    "max_samples": null,
    "total_samples": 4967824,
    "num_languages": 3,
    "samples_per_language": {
      "ell_Grek": 3745712,
      "kan_Knda": 677804,
      "swa_Latn": 544308
    }
  },
  "method": {
    "vocab_size": 261000,
    "base_training_method": "hf",
    "per_lang_counts_method": "soft",
    "byte_level": true,
    "seed": 42,
    "initial_vocab": "/workspace/hf_home/hub/models--CohereLabs--tiny-aya-global/snapshots/b2f40192d990d66a6f28a4b90d72817ab9c613da/tokenizer.json",
    "lang_batch_size": 5,
    "sampling": {
      "mode": "separate",
      "max_base_samples_per_lang": 10000,
      "max_lang_samples_per_lang": 100000,
      "shared_samples_per_lang": null
    },
    "reuse": {
      "reuse_corpus": true,
      "reuse_base": true,
      "skip_existing_langs": true
    }
  },
  "timing": {
    "total_seconds": 578.13,
    "base_tokenizer_seconds": 0.0,
    "language_tokenizers_seconds": 578.13,
    "base_tokenizer_reused": true
  },
  "output": {
    "results_dir": "/workspace/runpod/tinyAyaLid/results_tinyaya_soft_full",
    "corpus_dir": "/workspace/runpod/tinyAyaLid/data/glotlid_group_21",
    "tokenizers_dir": "/workspace/runpod/tinyAyaLid/results_tinyaya_soft_full/tokenizers",
    "base_tokenizer": "/workspace/runpod/tinyAyaLid/results_tinyaya_soft_full/tokenizers/langspec_base_tokenizer.json",
    "num_languages_trained_this_run": 3,
    "language_tokenizers": {
      "base_path": "results_tinyaya_soft_full/tokenizers/langspec_base_tokenizer.json",
      "language_paths": {
        "ell_Grek": "results_tinyaya_soft_full/tokenizers/langspec_soft_ell_Grek.tokenizer.json",
        "kan_Knda": "results_tinyaya_soft_full/tokenizers/langspec_soft_kan_Knda.tokenizer.json",
        "swa_Latn": "results_tinyaya_soft_full/tokenizers/langspec_soft_swa_Latn.tokenizer.json"
      }
    }
  },
  "languages": [
    "ell_Grek",
    "kan_Knda",
    "swa_Latn"
  ]
}