{
 "source_sha256": "f98299737e75bfb7002e47a76a67161246fbf1286c7be43f8ccd5800b50d6c17",
 "source_is_v2": true,
 "T1": {
  "table1_numbers_with_thousands": [
   6811,
   4564
  ],
  "text_totals": [
   {
    "line": 126,
    "text": "11,376 examples"
   },
   {
    "line": 168,
    "text": "11,376 examples"
   }
  ]
 },
 "T2": {
  "metrics": 12,
  "overall_printed": {
   "Char-RM": 0.631,
   "1-shot": 0.362,
   "2-shot": 0.385,
   "3-shot": 0.375
  },
  "mean_of_metrics": {
   "Char-RM": 0.5611,
   "1-shot": 0.3676,
   "2-shot": 0.3852,
   "3-shot": 0.3951
  },
  "gpt4_overall_increasing": false
 },
 "T4": {
  "models": [
   15,
   15
  ],
  "averages": 90,
  "max_abs_diff": 0.00067,
  "printed_not_equal_round3_of_mean": [
   {
    "table": "Table 4",
    "model": "ChatGLM3-6B",
    "column": "ca_avg",
    "printed": 3.399,
    "mean_of_printed_metrics": 3.39967
   },
   {
    "table": "Table 4",
    "model": "ChatGLM3-6B",
    "column": "ra_avg",
    "printed": 2.739,
    "mean_of_printed_metrics": 2.7395
   },
   {
    "table": "Table 4",
    "model": "InternLM-20B",
    "column": "ra_avg",
    "printed": 2.911,
    "mean_of_printed_metrics": 2.9115
   },
   {
    "table": "Table 4",
    "model": "BC-NPC-Turbo",
    "column": "ca_avg",
    "printed": 3.798,
    "mean_of_printed_metrics": 3.79733
   },
   {
    "table": "GPT-4 table",
    "model": "InternLM-7B",
    "column": "ca_avg",
    "printed": 4.418,
    "mean_of_printed_metrics": 4.41867
   },
   {
    "table": "GPT-4 table",
    "model": "Xingchen",
    "column": "ra_avg",
    "printed": 3.539,
    "mean_of_printed_metrics": 3.5385
   },
   {
    "table": "GPT-4 table",
    "model": "MiniMax",
    "column": "cc_avg",
    "printed": 4.763,
    "mean_of_printed_metrics": 4.7636
   },
   {
    "table": "GPT-4 table",
    "model": "MiniMax",
    "column": "ca_avg",
    "printed": 4.71,
    "mean_of_printed_metrics": 4.71067
   },
   {
    "table": "GPT-4 table",
    "model": "GPT-3.5",
    "column": "ra_avg",
    "printed": 3.23,
    "mean_of_printed_metrics": 3.2295
   }
  ],
  "bold_underline_mismatches": [
   {
    "column": "cons",
    "model": "Baichuan2-7B",
    "value": 3.827,
    "bold": false,
    "underline": true,
    "best": 3.916,
    "second": 3.864
   },
   {
    "column": "cons",
    "model": "Baichuan2-13B",
    "value": 3.864,
    "bold": false,
    "underline": false,
    "best": 3.916,
    "second": 3.864
   }
  ]
 },
 "MB": {
  "values": 14,
  "all_k_over_216": true,
  "k": {
   "ChatGLM3-6B": 115,
   "XVERSE-7B": 134,
   "Baichuan2-7B": 135,
   "Qwen-7B": 131,
   "InternLM-7B": 136,
   "XVERSE-13B": 136,
   "Baichuan2-13B": 138,
   "Qwen-14B": 134,
   "InternLM-20B": 140,
   "Xingchen": 136,
   "MiniMax": 148,
   "BC-NPC-Turbo": 147,
   "GPT-3.5": 141,
   "GPT-4": 150
  }
 },
 "NB": {
  "text": [
   {
    "line": 530,
    "text": "10 baselines"
   }
  ],
  "table4_rows": 15
 }
}