gen_stats.json

{
  "target_model": "meta-llama/Llama-3.2-1B-Instruct",
  "hf_config": "3.0.0",
  "frozen_prompt": {
    "system": "You are a helpful assistant that writes a brief, factual summary of a news article.",
    "user_template": "Summarize the following article:\n\n{article}",
    "add_generation_prompt": true
  },
  "decode": {
    "temperature": 0.0,
    "top_p": null,
    "max_new_tokens": 128,
    "greedy": true
  },
  "filter": {
    "max_model_len": 2048,
    "article_token_budget": 1837,
    "template_overhead": 59
  },
  "num_buckets": 8,
  "seed": 0,
  "dry_run": false,
  "splits": [
    {
      "source": "train",
      "written": 20000,
      "dropped_empty": 0,
      "dropped_degenerate": 0,
      "hit_max_new": 18561,
      "bucket_counts": {
        "0": 2500,
        "1": 2500,
        "2": 2500,
        "3": 2500,
        "4": 2500,
        "5": 2500,
        "6": 2500,
        "7": 2500
      },
      "summary_token_percentiles": {
        "p5": 121,
        "p25": 128,
        "p50": 128,
        "p75": 128,
        "p95": 128,
        "p100": 128
      },
      "path": "experiments/cnn_dailymail/data/train.jsonl"
    },
    {
      "source": "test",
      "written": 500,
      "dropped_empty": 0,
      "dropped_degenerate": 0,
      "hit_max_new": 445,
      "bucket_counts": {
        "0": 63,
        "1": 63,
        "2": 63,
        "3": 63,
        "4": 62,
        "5": 62,
        "6": 62,
        "7": 62
      },
      "summary_token_percentiles": {
        "p5": 114,
        "p25": 128,
        "p50": 128,
        "p75": 128,
        "p95": 128,
        "p100": 128
      },
      "path": "experiments/cnn_dailymail/data/test.jsonl"
    }
  ]
}