{
  "outputs": {
    "instruction-calibration.jsonl": {
      "counts": {
        "calibration/chat": {
          "languages": {
            "en": 90
          },
          "prediction_tokens": 90442,
          "records": 90,
          "tokens": 90532
        },
        "calibration/code": {
          "languages": {
            "en": 170
          },
          "prediction_tokens": 90007,
          "records": 170,
          "tokens": 90177
        },
        "calibration/multilingual": {
          "languages": {
            "acq": 1,
            "amh": 2,
            "arb": 8,
            "ary": 12,
            "arz": 1,
            "ben": 4,
            "ell": 2,
            "eng": 6,
            "eus": 3,
            "fil": 4,
            "fin": 3,
            "fra": 1,
            "guj": 11,
            "hau": 6,
            "hin": 1,
            "ibo": 3,
            "ita": 1,
            "jpn": 17,
            "kir": 14,
            "lit": 1,
            "mal": 5,
            "mar": 6,
            "nld": 4,
            "npi": 5,
            "nya": 1,
            "pan": 10,
            "pes": 2,
            "plt": 24,
            "por": 10,
            "rus": 1,
            "sin": 25,
            "sna": 1,
            "snd": 1,
            "som": 15,
            "spa": 4,
            "swe": 1,
            "swh": 1,
            "tam": 25,
            "tel": 14,
            "tur": 12,
            "ukr": 1,
            "urd": 4,
            "vie": 12,
            "wol": 4,
            "xho": 1,
            "yor": 15,
            "zho": 6,
            "zsm": 16,
            "zul": 4
          },
          "prediction_tokens": 89762,
          "records": 331,
          "tokens": 90093
        },
        "calibration/tools": {
          "languages": {
            "en": 23
          },
          "prediction_tokens": 30116,
          "records": 23,
          "tokens": 30139
        }
      },
      "max_record_tokens": 1979,
      "min_record_tokens": 32,
      "records": 614,
      "sha256": "c65c34c37633381a7322ff433a292f9e615d414bb834d1951b243ee16ecb310e",
      "tokens": 300941
    },
    "instruction-eval.jsonl": {
      "counts": {
        "heldout/chat": {
          "languages": {
            "en": 2
          },
          "prediction_tokens": 950,
          "records": 2,
          "tokens": 952
        },
        "heldout/code": {
          "languages": {
            "en": 4
          },
          "prediction_tokens": 953,
          "records": 4,
          "tokens": 957
        },
        "heldout/multilingual": {
          "languages": {
            "arb": 1,
            "eng": 2,
            "por": 1,
            "tel": 1,
            "tur": 1,
            "zho": 1
          },
          "prediction_tokens": 1010,
          "records": 7,
          "tokens": 1017
        },
        "heldout/tools": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 1020,
          "records": 1,
          "tokens": 1021
        },
        "validation/chat": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 994,
          "records": 1,
          "tokens": 995
        },
        "validation/code": {
          "languages": {
            "en": 3
          },
          "prediction_tokens": 914,
          "records": 3,
          "tokens": 917
        },
        "validation/multilingual": {
          "languages": {
            "arb": 1,
            "eng": 1,
            "por": 1,
            "yor": 1
          },
          "prediction_tokens": 963,
          "records": 4,
          "tokens": 967
        },
        "validation/tools": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 937,
          "records": 1,
          "tokens": 938
        }
      },
      "max_record_tokens": 1021,
      "min_record_tokens": 38,
      "records": 23,
      "sha256": "a53a6e18c4d33b787f15bb1aad6ce88de97f25239ad2236617a62256b297f979",
      "tokens": 7764
    },
    "instruction-long-context.jsonl": {
      "counts": {
        "heldout/chat": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 2241,
          "records": 1,
          "tokens": 2242
        },
        "heldout/multilingual": {
          "languages": {
            "arb": 1
          },
          "prediction_tokens": 2294,
          "records": 1,
          "tokens": 2295
        },
        "heldout/tools": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 2829,
          "records": 1,
          "tokens": 2830
        },
        "validation/chat": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 2377,
          "records": 1,
          "tokens": 2378
        },
        "validation/multilingual": {
          "languages": {
            "arb": 1
          },
          "prediction_tokens": 2475,
          "records": 1,
          "tokens": 2476
        },
        "validation/tools": {
          "languages": {
            "en": 1
          },
          "prediction_tokens": 2331,
          "records": 1,
          "tokens": 2332
        }
      },
      "max_record_tokens": 2830,
      "min_record_tokens": 2242,
      "records": 6,
      "sha256": "6ca2b86259d40fe30ab5db341afec97f9a0efd8074f0754f8978988b0d3f6d3f",
      "tokens": 14553
    }
  },
  "packages": {
    "jinja2": "3.1.6",
    "tokenizers": "0.22.2",
    "transformers": "5.12.1"
  },
  "purpose": "Diverse exact-template instruct PTQ calibration for TT sensitivity/precision ranking; not official Unsloth data or a llama.cpp imatrix; no training/QAT/QAD",
  "rejected_candidates": {
    "chat:outside_32_4096": 2,
    "multilingual:normalized_prompt_duplicate": 1,
    "multilingual:outside_32_4096": 116,
    "tools:conversion:Expecting value: line 1 column 1 (char 0)": 326,
    "tools:conversion:No real tool schemas": 24,
    "tools:conversion:Require a complete user-to-assistant conversation": 1,
    "tools:conversion:Tool response/call order mismatch": 63
  },
  "schema_version": 1,
  "script_sha256": "c5abd94571e74b281cff61c342c5037742e5384ed3a478ce6c9c3df7d42fc5fa",
  "selection": {
    "calibration_target_tokens": {
      "chat": 90000,
      "code": 90000,
      "multilingual": 90000,
      "tools": 30000
    },
    "candidate_order": "ascending SHA256(id + ':order-v1')",
    "context_policy": "No truncation or packing. Entire conversations only; 32..2048 calibration, 32..1024 small KL, 2049..4096 separate long contexts. Oversized or malformed/truncated source rows rejected.",
    "dedup": "Global exact normalized user-prompt SHA256 (Unicode NFKC, casefold, whitespace collapse), across all domains and all splits including long contexts; reject entire conversation if any user prompt already used. Semantic/paraphrase dedup not claimed.",
    "eval_independence": "No calibration prompt overlap; official heldout source splits where available, deterministic disjoint row partitions otherwise. Same dataset families, NOT independent-source generalization proof.",
    "kl_total_token_cap": 8192,
    "multilingual_eval_order": "Stable language-round-robin over hash order: first occurrence of every language before second occurrences, etc.",
    "source_rows": "50-row windows at evenly spaced integer offsets spanning each entire source split; source order preserved inside each window",
    "split_assignment": "Official train/test for UltraChat and Aya; test hashed 50/50 validation/heldout. Code/Hermes train rows hashed 80/10/10. SHA256(domain:source_split:row_idx:split-v1) first8 hex mod10."
  },
  "snapshot_manifest_sha256": "d73630cfe8466e5545fdeba82f9dd342a9ea5d4232facfe94f93f4119a9f5b58",
  "sources": {
    "chat": {
      "config": "default",
      "dataset": "HuggingFaceH4/ultrachat_200k",
      "license": "mit",
      "revision": "8049631c405ae6576f93f445c6b8166f76f5505a",
      "splits": {
        "test_sft": 10,
        "train_sft": 20
      }
    },
    "code": {
      "config": "default",
      "dataset": "m-a-p/CodeFeedback-Filtered-Instruction",
      "license": "apache-2.0",
      "revision": "a08c213a9748c66c15d0225814be80a2e77adf4a",
      "splits": {
        "train": 24
      }
    },
    "multilingual": {
      "config": "default",
      "dataset": "CohereLabs/aya_dataset",
      "license": "apache-2.0",
      "revision": "f9ea04583f02a8f86404ff6c58bf75fe637df8a2",
      "splits": {
        "test": 10,
        "train": 40
      }
    },
    "tools": {
      "config": "func_calling",
      "dataset": "NousResearch/hermes-function-calling-v1",
      "license": "apache-2.0",
      "revision": "dae3e1d28cfbcf4b915c04ea1e072030529b4bda",
      "splits": {
        "train": 16
      }
    }
  },
  "tokenizer": {
    "add_generation_prompt": false,
    "chat_template_applied": true,
    "enable_thinking": false,
    "files": {
      "chat_template.jinja": "a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715",
      "tokenizer.json": "5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42",
      "tokenizer_config.json": "316230d6a809701f4db5ea8f8fc862bc3a6f3229c937c174e674ff3ca0a64ac8"
    },
    "model": "Qwen/Qwen3.5-9B",
    "revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
    "tools": "Real Hermes function schemas, assistant.tool_calls argument objects and linked tool responses; source system tool boilerplate replaced by exact Qwen template tools rendering"
  }
}