{
 "schema_version": "1.0",
 "generated_at": "2026-10-21T15:00:02+00:00",
 "contains_example_data": false,
 "notice": "All rows are real measurements; raw logs at /logs/<run_id>/.",
 "models": [
  {
   "id": "granite-4.2-8b",
   "name": "Granite 4.2 8B",
   "family": "Granite",
   "params_b": 8.79,
   "active_params_b": null,
   "base_repo": "ibm-granite/granite-4.2-8b",
   "reference_dtype": "BF16",
   "reference_gguf": null,
   "context_length": 131072,
   "kv_bytes_per_token": 163840,
   "description": "ibm-granite/granite-4.2-8b (Apache-2.0). Dense 40-layer transformer with grouped-query attention (8 KV heads). Measured here: on the 60 parallel and parallel_multiple tool-call cases it emits only one call per turn, so all 8 quants scored 0 to 2 of 60 there (Q8_0: 0/60, bartowski Q4_K_M: 1/60), even with parallel_tool_calls=true. Its total tool-call score is 60.5 to 62.5% vs 82 to 92.5% for Qwen3.5 9B.",
   "example": false,
   "license": "apache-2.0"
  },
  {
   "id": "qwen3.5-4b",
   "name": "Qwen3.5 4B",
   "family": "Qwen3.5",
   "params_b": 4.21,
   "active_params_b": null,
   "base_repo": "Qwen/Qwen3.5-4B",
   "reference_dtype": "BF16",
   "reference_gguf": null,
   "context_length": 262144,
   "kv_bytes_per_token": 32768,
   "description": "Qwen/Qwen3.5-4B (Apache-2.0). Same hybrid design as the 9B (8 of 32 layers full attention). Text-only GGUF quants. Tool calls tested with thinking disabled.",
   "example": false,
   "license": "apache-2.0"
  },
  {
   "id": "qwen3.5-9b",
   "name": "Qwen3.5 9B",
   "family": "Qwen3.5",
   "params_b": 8.95,
   "active_params_b": null,
   "base_repo": "Qwen/Qwen3.5-9B",
   "reference_dtype": "BF16",
   "reference_gguf": null,
   "context_length": 262144,
   "kv_bytes_per_token": 32768,
   "description": "Qwen/Qwen3.5-9B (Apache-2.0). Hybrid architecture: 32 layers, of which 8 use full attention and 24 linear attention, so the KV cache is small. The GGUF quants are text-only (vision tower not included). Tool calls tested with thinking disabled.",
   "example": false,
   "license": "apache-2.0"
  }
 ],
 "quants": [
  {
   "id": "granite-4.2-8b--bartowski--iq3_m",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-IQ3_M.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "2d2b4dac668a78751b90912ddea1157587ef4657362ff8e13ca948a98f780dd8",
   "format": "GGUF",
   "quant_type": "IQ3_M",
   "imatrix": true,
   "size_bytes": 4241228192,
   "bpw": 3.86,
   "example": false
  },
  {
   "id": "granite-4.2-8b--bartowski--iq4_xs",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-IQ4_XS.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "729d7338f221e3f080a43a0c29ba0ea4b862e2c4b08c52d85426022c0cf5e71b",
   "format": "GGUF",
   "quant_type": "IQ4_XS",
   "imatrix": true,
   "size_bytes": 4939563424,
   "bpw": 4.49,
   "example": false
  },
  {
   "id": "granite-4.2-8b--bartowski--q2_k",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-Q2_K.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "a465c49d0000029d09d1de7bd8e37b9d7848dfc39d6cf1aa450a4e7162c3160c",
   "format": "GGUF",
   "quant_type": "Q2_K",
   "imatrix": true,
   "size_bytes": 3577463200,
   "bpw": 3.26,
   "example": false
  },
  {
   "id": "granite-4.2-8b--bartowski--q4_k_m",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-Q4_K_M.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "142c0fe97ec32d9c9cbc66460761e1d3375ee679f60a6534ae0fd3189cc7f5ea",
   "format": "GGUF",
   "quant_type": "Q4_K_M",
   "imatrix": true,
   "size_bytes": 5539283360,
   "bpw": 5.04,
   "example": false
  },
  {
   "id": "granite-4.2-8b--bartowski--q6_k",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-Q6_K.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "f39edecda3cfc42f2448caa44c4b1753dbf511ccdf4eb18b09f1e9f5922f900d",
   "format": "GGUF",
   "quant_type": "Q6_K",
   "imatrix": true,
   "size_bytes": 7521223072,
   "bpw": 6.84,
   "example": false
  },
  {
   "id": "granite-4.2-8b--bartowski--q8_0",
   "model_id": "granite-4.2-8b",
   "publisher": "bartowski",
   "repo": "bartowski/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b-Q8_0.gguf",
   "revision": "a592100df8fe4931c7cffbac7b28e8176a1d52da",
   "sha256": "b06eac4b12ee3b65a8bf9c6fd122fe46203291fd9c09cba7765d4d6400419aa9",
   "format": "GGUF",
   "quant_type": "Q8_0",
   "imatrix": true,
   "size_bytes": 9345614240,
   "bpw": 8.5,
   "example": false
  },
  {
   "id": "granite-4.2-8b--mradermacher--i1-q4_k_m",
   "model_id": "granite-4.2-8b",
   "publisher": "mradermacher",
   "repo": "mradermacher/granite-4.2-8b-i1-GGUF",
   "filename": "granite-4.2-8b.i1-Q4_K_M.gguf",
   "revision": "4e036f43e7e72a464093de36e41388a5b1d35076",
   "sha256": "78174205d4ac501a1e79337c46cef73e1c9085d1b61cf41e74820d3214ee3a38",
   "format": "GGUF",
   "quant_type": "i1-Q4_K_M",
   "imatrix": true,
   "size_bytes": 5347918624,
   "bpw": 4.87,
   "example": false
  },
  {
   "id": "granite-4.2-8b--mradermacher--q5_k_m",
   "model_id": "granite-4.2-8b",
   "publisher": "mradermacher",
   "repo": "mradermacher/granite-4.2-8b-GGUF",
   "filename": "granite-4.2-8b.Q5_K_M.gguf",
   "revision": "72e0876f143f0642399a9a0b2565b697706d7c27",
   "sha256": "ace0997c84444984c849080168964978d16704a79f9aaa8e3d6dc116b479923e",
   "format": "GGUF",
   "quant_type": "Q5_K_M",
   "imatrix": false,
   "size_bytes": 6253888064,
   "bpw": 5.69,
   "example": false
  },
  {
   "id": "qwen3.5-4b--bartowski--iq3_m",
   "model_id": "qwen3.5-4b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-4B-GGUF",
   "filename": "Qwen_Qwen3.5-4B-IQ3_M.gguf",
   "revision": "4168f45a16a1290d65a4ec0fa312ae917a4c15d6",
   "sha256": "5ad76abae6b7a27d1463d90e6d86886850769e34906a88cfa4347cae72c91502",
   "format": "GGUF",
   "quant_type": "IQ3_M",
   "imatrix": true,
   "size_bytes": 2483333088,
   "bpw": 4.59,
   "example": false
  },
  {
   "id": "qwen3.5-4b--bartowski--q4_k_m",
   "model_id": "qwen3.5-4b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-4B-GGUF",
   "filename": "Qwen_Qwen3.5-4B-Q4_K_M.gguf",
   "revision": "4168f45a16a1290d65a4ec0fa312ae917a4c15d6",
   "sha256": "13c16f426047e2de38cd075bdade4a7bcbc8c774384876f677740cda65f8a983",
   "format": "GGUF",
   "quant_type": "Q4_K_M",
   "imatrix": true,
   "size_bytes": 3013027808,
   "bpw": 5.57,
   "example": false
  },
  {
   "id": "qwen3.5-4b--bartowski--q6_k",
   "model_id": "qwen3.5-4b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-4B-GGUF",
   "filename": "Qwen_Qwen3.5-4B-Q6_K.gguf",
   "revision": "4168f45a16a1290d65a4ec0fa312ae917a4c15d6",
   "sha256": "5a5bc7a3f9375b395e9f81fb69109038cf160c84606b8192f13cbb8355ca59f2",
   "format": "GGUF",
   "quant_type": "Q6_K",
   "imatrix": true,
   "size_bytes": 3805358048,
   "bpw": 7.04,
   "example": false
  },
  {
   "id": "qwen3.5-4b--mradermacher--i1-q4_k_m",
   "model_id": "qwen3.5-4b",
   "publisher": "mradermacher",
   "repo": "mradermacher/Qwen3.5-4B-i1-GGUF",
   "filename": "Qwen3.5-4B.i1-Q4_K_M.gguf",
   "revision": "8501a8b9bf0a85a86e4abce7b17870bcfad5d128",
   "sha256": "b373d4f4179d75d35a74170f058bc4d2136edbedd536e90b9aa01bda76ef96ef",
   "format": "GGUF",
   "quant_type": "i1-Q4_K_M",
   "imatrix": true,
   "size_bytes": 2708805056,
   "bpw": 5.15,
   "example": false
  },
  {
   "id": "qwen3.5-4b--unsloth--q4_k_m",
   "model_id": "qwen3.5-4b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-4B-GGUF",
   "filename": "Qwen3.5-4B-Q4_K_M.gguf",
   "revision": "e87f176479d0855a907a41277aca2f8ee7a09523",
   "sha256": "00fe7986ff5f6b463e62455821146049db6f9313603938a70800d1fb69ef11a4",
   "format": "GGUF",
   "quant_type": "Q4_K_M",
   "imatrix": null,
   "size_bytes": 2740937888,
   "bpw": 5.21,
   "example": false
  },
  {
   "id": "qwen3.5-4b--unsloth--q8_0",
   "model_id": "qwen3.5-4b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-4B-GGUF",
   "filename": "Qwen3.5-4B-Q8_0.gguf",
   "revision": "e87f176479d0855a907a41277aca2f8ee7a09523",
   "sha256": "10cc391b403021dd11c614679d2fd92f611c3681d29e29651b717316965d61e1",
   "format": "GGUF",
   "quant_type": "Q8_0",
   "imatrix": null,
   "size_bytes": 4482403488,
   "bpw": 8.53,
   "example": false
  },
  {
   "id": "qwen3.5-4b--unsloth--ud-q2_k_xl",
   "model_id": "qwen3.5-4b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-4B-GGUF",
   "filename": "Qwen3.5-4B-UD-Q2_K_XL.gguf",
   "revision": "e87f176479d0855a907a41277aca2f8ee7a09523",
   "sha256": "79aa0b583c888976013002f66b9f617070d91cf8de6a09dc775983640ac59463",
   "format": "GGUF",
   "quant_type": "UD-Q2_K_XL",
   "imatrix": true,
   "size_bytes": 1940825248,
   "bpw": 3.69,
   "example": false
  },
  {
   "id": "qwen3.5-4b--unsloth--ud-q4_k_xl",
   "model_id": "qwen3.5-4b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-4B-GGUF",
   "filename": "Qwen3.5-4B-UD-Q4_K_XL.gguf",
   "revision": "e87f176479d0855a907a41277aca2f8ee7a09523",
   "sha256": "b252c5610a42ca82d20fe2a12813e9d069eed89292907e26c783eeb0bc961bc7",
   "format": "GGUF",
   "quant_type": "UD-Q4_K_XL",
   "imatrix": true,
   "size_bytes": 2912109728,
   "bpw": 5.54,
   "example": false
  },
  {
   "id": "qwen3.5-9b--bartowski--iq3_m",
   "model_id": "qwen3.5-9b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-9B-GGUF",
   "filename": "Qwen_Qwen3.5-9B-IQ3_M.gguf",
   "revision": "182be2fd6c7bc44887d88a91cb03ff009cc9f549",
   "sha256": "cbeadcf49deaf24c756516c4b98e88b88ec8719108553038623d52907f84522d",
   "format": "GGUF",
   "quant_type": "IQ3_M",
   "imatrix": true,
   "size_bytes": 4981092384,
   "bpw": 4.33,
   "example": false
  },
  {
   "id": "qwen3.5-9b--bartowski--iq4_xs",
   "model_id": "qwen3.5-9b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-9B-GGUF",
   "filename": "Qwen_Qwen3.5-9B-IQ4_XS.gguf",
   "revision": "182be2fd6c7bc44887d88a91cb03ff009cc9f549",
   "sha256": "7d977cc96c2e08616016d967f232083e354691a8a16f345b26f7d782ee5c9601",
   "format": "GGUF",
   "quant_type": "IQ4_XS",
   "imatrix": true,
   "size_bytes": 5501202464,
   "bpw": 4.79,
   "example": false
  },
  {
   "id": "qwen3.5-9b--bartowski--q4_k_m",
   "model_id": "qwen3.5-9b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-9B-GGUF",
   "filename": "Qwen_Qwen3.5-9B-Q4_K_M.gguf",
   "revision": "182be2fd6c7bc44887d88a91cb03ff009cc9f549",
   "sha256": "d784ce9eda1a5a7b51e8f705a9e6310844bf4f173654d115823c775fdea56d43",
   "format": "GGUF",
   "quant_type": "Q4_K_M",
   "imatrix": true,
   "size_bytes": 6169341984,
   "bpw": 5.37,
   "example": false
  },
  {
   "id": "qwen3.5-9b--bartowski--q5_k_m",
   "model_id": "qwen3.5-9b",
   "publisher": "bartowski",
   "repo": "bartowski/Qwen_Qwen3.5-9B-GGUF",
   "filename": "Qwen_Qwen3.5-9B-Q5_K_M.gguf",
   "revision": "182be2fd6c7bc44887d88a91cb03ff009cc9f549",
   "sha256": "a686d88ec1e6881f9bf161526826cd6d6874b7f0e80e0f79acf6144a132c5d7e",
   "format": "GGUF",
   "quant_type": "Q5_K_M",
   "imatrix": true,
   "size_bytes": 7111487520,
   "bpw": 6.19,
   "example": false
  },
  {
   "id": "qwen3.5-9b--mradermacher--i1-iq4_xs",
   "model_id": "qwen3.5-9b",
   "publisher": "mradermacher",
   "repo": "mradermacher/Qwen3.5-9B-i1-GGUF",
   "filename": "Qwen3.5-9B.i1-IQ4_XS.gguf",
   "revision": "c5044c600a7e15246d54d13e4532382bc44ef41b",
   "sha256": "95ed3f9b8c28571d3aa5c4a426bd16664f0f7bdc5ff14a1438c5e9dcace6979d",
   "format": "GGUF",
   "quant_type": "i1-IQ4_XS",
   "imatrix": true,
   "size_bytes": 5070611968,
   "bpw": 4.53,
   "example": false
  },
  {
   "id": "qwen3.5-9b--mradermacher--i1-q4_k_m",
   "model_id": "qwen3.5-9b",
   "publisher": "mradermacher",
   "repo": "mradermacher/Qwen3.5-9B-i1-GGUF",
   "filename": "Qwen3.5-9B.i1-Q4_K_M.gguf",
   "revision": "c5044c600a7e15246d54d13e4532382bc44ef41b",
   "sha256": "611d3302fc059d0f1d9b72e3e25e389aa8549d71e1081b50b3dc0c6b99314741",
   "format": "GGUF",
   "quant_type": "i1-Q4_K_M",
   "imatrix": true,
   "size_bytes": 5627045376,
   "bpw": 5.03,
   "example": false
  },
  {
   "id": "qwen3.5-9b--unsloth--q4_k_m",
   "model_id": "qwen3.5-9b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-9B-GGUF",
   "filename": "Qwen3.5-9B-Q4_K_M.gguf",
   "revision": "3885219b6810b007914f3a7950a8d1b469d598a5",
   "sha256": "03b74727a860a56338e042c4420bb3f04b2fec5734175f4cb9fa853daf52b7e8",
   "format": "GGUF",
   "quant_type": "Q4_K_M",
   "imatrix": null,
   "size_bytes": 5680522464,
   "bpw": 5.08,
   "example": false
  },
  {
   "id": "qwen3.5-9b--unsloth--q6_k",
   "model_id": "qwen3.5-9b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-9B-GGUF",
   "filename": "Qwen3.5-9B-Q6_K.gguf",
   "revision": "3885219b6810b007914f3a7950a8d1b469d598a5",
   "sha256": "91898433cf5ce0a8f45516a4cc3e9343b6e01d052d01f684309098c66a326c59",
   "format": "GGUF",
   "quant_type": "Q6_K",
   "imatrix": null,
   "size_bytes": 7458301152,
   "bpw": 6.66,
   "example": false
  },
  {
   "id": "qwen3.5-9b--unsloth--q8_0",
   "model_id": "qwen3.5-9b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-9B-GGUF",
   "filename": "Qwen3.5-9B-Q8_0.gguf",
   "revision": "3885219b6810b007914f3a7950a8d1b469d598a5",
   "sha256": "809626574d0cb43d4becfa56169980da2bb448f2299270f7be443cb89d0a6ae4",
   "format": "GGUF",
   "quant_type": "Q8_0",
   "imatrix": null,
   "size_bytes": 9527502048,
   "bpw": 8.51,
   "example": false
  },
  {
   "id": "qwen3.5-9b--unsloth--ud-q2_k_xl",
   "model_id": "qwen3.5-9b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-9B-GGUF",
   "filename": "Qwen3.5-9B-UD-Q2_K_XL.gguf",
   "revision": "3885219b6810b007914f3a7950a8d1b469d598a5",
   "sha256": "fcea605227c939525a4145349a082f603d7e5df639be3d88b7aa1e883517552c",
   "format": "GGUF",
   "quant_type": "UD-Q2_K_XL",
   "imatrix": true,
   "size_bytes": 4121781472,
   "bpw": 3.68,
   "example": false
  },
  {
   "id": "qwen3.5-9b--unsloth--ud-q4_k_xl",
   "model_id": "qwen3.5-9b",
   "publisher": "unsloth",
   "repo": "unsloth/Qwen3.5-9B-GGUF",
   "filename": "Qwen3.5-9B-UD-Q4_K_XL.gguf",
   "revision": "3885219b6810b007914f3a7950a8d1b469d598a5",
   "sha256": "6f5d30666c2d8ae16a306e616d95341dcf3cc46810df84d7e6f5a7d1e4c1b293",
   "format": "GGUF",
   "quant_type": "UD-Q4_K_XL",
   "imatrix": true,
   "size_bytes": 5966095584,
   "bpw": 5.33,
   "example": false
  }
 ],
 "runs": [
  {
   "id": "20261021T073431_granite-4.2-8b--bartowski--iq3_m",
   "quant_id": "granite-4.2-8b--bartowski--iq3_m",
   "started_at": "2026-10-21T07:34:31Z",
   "finished_at": "2026-10-21T07:40:21Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ3_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ3_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ3_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.223337,
     "max": 16.911457,
     "p99": 2.836165,
     "median": 0.068018,
     "same_top_pct": 82.925,
     "ppl": 13.414878,
     "ppl_ref": 12.811159,
     "delta_p_rms": 12.135
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4401.61,
     "pp_stddev": 96.04,
     "tg_tps": 119.63,
     "tg_stddev": 0.59
    },
    "toolcall": {
     "score": 60.5,
     "passed": 121,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 29,
       "total": 40,
       "score": 72.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T070546_granite-4.2-8b--bartowski--iq4_xs",
   "quant_id": "granite-4.2-8b--bartowski--iq4_xs",
   "started_at": "2026-10-21T07:05:46Z",
   "finished_at": "2026-10-21T07:13:20Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ4_XS.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ4_XS.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-IQ4_XS.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.084645,
     "max": 20.52557,
     "p99": 1.0248,
     "median": 0.020277,
     "same_top_pct": 89.541,
     "ppl": 12.718055,
     "ppl_ref": 12.811159,
     "delta_p_rms": 7.495
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4807.65,
     "pp_stddev": 127.29,
     "tg_tps": 112.55,
     "tg_stddev": 0.54
    },
    "toolcall": {
     "score": 62.5,
     "passed": 125,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 55,
       "total": 60,
       "score": 91.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 34,
       "total": 40,
       "score": 85.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T074021_granite-4.2-8b--bartowski--q2_k",
   "quant_id": "granite-4.2-8b--bartowski--q2_k",
   "started_at": "2026-10-21T07:40:21Z",
   "finished_at": "2026-10-21T07:46:09Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q2_K.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q2_K.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q2_K.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.459788,
     "max": 21.899715,
     "p99": 5.202459,
     "median": 0.168584,
     "same_top_pct": 74.961,
     "ppl": 13.879087,
     "ppl_ref": 12.811159,
     "delta_p_rms": 17.378
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 3315.22,
     "pp_stddev": 63.63,
     "tg_tps": 145.65,
     "tg_stddev": 0.3
    },
    "toolcall": {
     "score": 61.0,
     "passed": 122,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 28,
       "total": 40,
       "score": 70.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T064958_granite-4.2-8b--bartowski--q4_k_m",
   "quant_id": "granite-4.2-8b--bartowski--q4_k_m",
   "started_at": "2026-10-21T06:49:58Z",
   "finished_at": "2026-10-21T06:59:02Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true,
    "toolcall_rerun_at": "2026-10-21T08:01:52Z",
    "toolcall_harness_version": "0.1.2"
   },
   "results": {
    "kld": {
     "mean": 0.072943,
     "max": 18.713558,
     "p99": 0.928473,
     "median": 0.014506,
     "same_top_pct": 91.0,
     "ppl": 13.049144,
     "ppl_ref": 12.811159,
     "delta_p_rms": 6.61
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4236.45,
     "pp_stddev": 97.64,
     "tg_tps": 101.61,
     "tg_stddev": 0.23
    },
    "toolcall": {
     "score": 62.5,
     "passed": 125,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 38,
       "total": 40,
       "score": 95.0
      },
      "parallel": {
       "passed": 1,
       "total": 30,
       "score": 3.3
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 30,
       "total": 40,
       "score": 75.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl",
    "toolcall_v0_1_1": "toolcall.v0.1.1.jsonl"
   },
   "example": false,
   "notes": "tool-call stage re-run 2026-10-21T08:01:52Z with parallel_tool_calls=true (first pass used llama-server's default, which allows at most one call per turn; old score 62.5%)"
  },
  {
   "id": "20261021T072036_granite-4.2-8b--bartowski--q6_k",
   "quant_id": "granite-4.2-8b--bartowski--q6_k",
   "started_at": "2026-10-21T07:20:36Z",
   "finished_at": "2026-10-21T07:27:52Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q6_K.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q6_K.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q6_K.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.0136,
     "max": 11.952006,
     "p99": 0.140037,
     "median": 0.002261,
     "same_top_pct": 96.227,
     "ppl": 12.952515,
     "ppl_ref": 12.811159,
     "delta_p_rms": 3.049
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 3678.92,
     "pp_stddev": 70.84,
     "tg_tps": 78.45,
     "tg_stddev": 0.36
    },
    "toolcall": {
     "score": 60.5,
     "passed": 121,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 29,
       "total": 40,
       "score": 72.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T071320_granite-4.2-8b--bartowski--q8_0",
   "quant_id": "granite-4.2-8b--bartowski--q8_0",
   "started_at": "2026-10-21T07:13:20Z",
   "finished_at": "2026-10-21T07:20:35Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q8_0.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q8_0.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--granite-4.2-8b-GGUF\\snapshots\\a592100df8fe4931c7cffbac7b28e8176a1d52da\\granite-4.2-8b-Q8_0.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.008183,
     "max": 12.706347,
     "p99": 0.062564,
     "median": 0.000778,
     "same_top_pct": 97.824,
     "ppl": 12.946332,
     "ppl_ref": 12.811159,
     "delta_p_rms": 1.977
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4574.16,
     "pp_stddev": 101.28,
     "tg_tps": 64.85,
     "tg_stddev": 0.17
    },
    "toolcall": {
     "score": 61.0,
     "passed": 122,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 29,
       "total": 40,
       "score": 72.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T065903_granite-4.2-8b--mradermacher--i1-q4_k_m",
   "quant_id": "granite-4.2-8b--mradermacher--i1-q4_k_m",
   "started_at": "2026-10-21T06:59:03Z",
   "finished_at": "2026-10-21T07:05:46Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-i1-GGUF\\snapshots\\4e036f43e7e72a464093de36e41388a5b1d35076\\granite-4.2-8b.i1-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-i1-GGUF\\snapshots\\4e036f43e7e72a464093de36e41388a5b1d35076\\granite-4.2-8b.i1-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-i1-GGUF\\snapshots\\4e036f43e7e72a464093de36e41388a5b1d35076\\granite-4.2-8b.i1-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.085565,
     "max": 15.774079,
     "p99": 1.188092,
     "median": 0.01741,
     "same_top_pct": 90.227,
     "ppl": 13.236764,
     "ppl_ref": 12.811159,
     "delta_p_rms": 7.028
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4247.62,
     "pp_stddev": 94.39,
     "tg_tps": 104.8,
     "tg_stddev": 0.68
    },
    "toolcall": {
     "score": 60.5,
     "passed": 121,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 1,
       "total": 30,
       "score": 3.3
      },
      "parallel_multiple": {
       "passed": 0,
       "total": 30,
       "score": 0.0
      },
      "irrelevance": {
       "passed": 29,
       "total": 40,
       "score": 72.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T072752_granite-4.2-8b--mradermacher--q5_k_m",
   "quant_id": "granite-4.2-8b--mradermacher--q5_k_m",
   "started_at": "2026-10-21T07:27:52Z",
   "finished_at": "2026-10-21T07:34:31Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-GGUF\\snapshots\\72e0876f143f0642399a9a0b2565b697706d7c27\\granite-4.2-8b.Q5_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\granite-4.2-8b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-GGUF\\snapshots\\72e0876f143f0642399a9a0b2565b697706d7c27\\granite-4.2-8b.Q5_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--granite-4.2-8b-GGUF\\snapshots\\72e0876f143f0642399a9a0b2565b697706d7c27\\granite-4.2-8b.Q5_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "ibm-granite/granite-4.2-8b -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 12.9206,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.057894,
     "max": 17.293224,
     "p99": 0.74641,
     "median": 0.011605,
     "same_top_pct": 92.059,
     "ppl": 13.22346,
     "ppl_ref": 12.811159,
     "delta_p_rms": 5.909
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4220.21,
     "pp_stddev": 93.05,
     "tg_tps": 92.29,
     "tg_stddev": 0.18
    },
    "toolcall": {
     "score": 62.5,
     "passed": 125,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 1,
       "total": 30,
       "score": 3.3
      },
      "parallel_multiple": {
       "passed": 1,
       "total": 30,
       "score": 3.3
      },
      "irrelevance": {
       "passed": 29,
       "total": 40,
       "score": 72.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T084433_qwen3.5-4b--bartowski--iq3_m",
   "quant_id": "qwen3.5-4b--bartowski--iq3_m",
   "started_at": "2026-10-21T08:44:33Z",
   "finished_at": "2026-10-21T08:48:31Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-IQ3_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-IQ3_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-IQ3_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.100231,
     "max": 12.282552,
     "p99": 0.929547,
     "median": 0.05189,
     "same_top_pct": 86.314,
     "ppl": 9.597882,
     "ppl_ref": 8.863479,
     "delta_p_rms": 9.198
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 8197.24,
     "pp_stddev": 402.98,
     "tg_tps": 161.39,
     "tg_stddev": 0.3
    },
    "toolcall": {
     "score": 87.5,
     "passed": 175,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "parallel_multiple": {
       "passed": 24,
       "total": 30,
       "score": 80.0
      },
      "irrelevance": {
       "passed": 31,
       "total": 40,
       "score": 77.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T081756_qwen3.5-4b--bartowski--q4_k_m",
   "quant_id": "qwen3.5-4b--bartowski--q4_k_m",
   "started_at": "2026-10-21T08:17:56Z",
   "finished_at": "2026-10-21T08:22:13Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.025793,
     "max": 10.865278,
     "p99": 0.213138,
     "median": 0.010966,
     "same_top_pct": 92.816,
     "ppl": 9.065974,
     "ppl_ref": 8.863479,
     "delta_p_rms": 4.48
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 7774.47,
     "pp_stddev": 404.86,
     "tg_tps": 151.77,
     "tg_stddev": 0.31
    },
    "toolcall": {
     "score": 86.5,
     "passed": 173,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 55,
       "total": 60,
       "score": 91.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 23,
       "total": 30,
       "score": 76.7
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T083951_qwen3.5-4b--bartowski--q6_k",
   "quant_id": "qwen3.5-4b--bartowski--q6_k",
   "started_at": "2026-10-21T08:39:51Z",
   "finished_at": "2026-10-21T08:44:33Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q6_K.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q6_K.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-Q6_K.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.003702,
     "max": 4.999385,
     "p99": 0.02415,
     "median": 0.001599,
     "same_top_pct": 97.106,
     "ppl": 8.923529,
     "ppl_ref": 8.863479,
     "delta_p_rms": 1.841
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 7187.83,
     "pp_stddev": 298.07,
     "tg_tps": 129.86,
     "tg_stddev": 0.17
    },
    "toolcall": {
     "score": 86.5,
     "passed": 173,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 55,
       "total": 60,
       "score": 91.7
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 24,
       "total": 30,
       "score": 80.0
      },
      "parallel_multiple": {
       "passed": 25,
       "total": 30,
       "score": 83.3
      },
      "irrelevance": {
       "passed": 34,
       "total": 40,
       "score": 85.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T082213_qwen3.5-4b--mradermacher--i1-q4_k_m",
   "quant_id": "qwen3.5-4b--mradermacher--i1-q4_k_m",
   "started_at": "2026-10-21T08:22:13Z",
   "finished_at": "2026-10-21T08:27:23Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-4B-i1-GGUF\\snapshots\\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\\Qwen3.5-4B.i1-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-4B-i1-GGUF\\snapshots\\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\\Qwen3.5-4B.i1-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-4B-i1-GGUF\\snapshots\\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\\Qwen3.5-4B.i1-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.033885,
     "max": 12.264156,
     "p99": 0.313766,
     "median": 0.014821,
     "same_top_pct": 91.847,
     "ppl": 9.050659,
     "ppl_ref": 8.863479,
     "delta_p_rms": 5.03
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 7785.33,
     "pp_stddev": 345.82,
     "tg_tps": 157.93,
     "tg_stddev": 0.41
    },
    "toolcall": {
     "score": 89.5,
     "passed": 179,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 27,
       "total": 30,
       "score": 90.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T081235_qwen3.5-4b--unsloth--q4_k_m",
   "quant_id": "qwen3.5-4b--unsloth--q4_k_m",
   "started_at": "2026-10-21T08:12:35Z",
   "finished_at": "2026-10-21T08:17:56Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.030754,
     "max": 11.893332,
     "p99": 0.250512,
     "median": 0.013181,
     "same_top_pct": 92.384,
     "ppl": 9.052141,
     "ppl_ref": 8.863479,
     "delta_p_rms": 4.766
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 7916.74,
     "pp_stddev": 357.48,
     "tg_tps": 156.35,
     "tg_stddev": 0.18
    },
    "toolcall": {
     "score": 87.0,
     "passed": 174,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 56,
       "total": 60,
       "score": 93.3
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 25,
       "total": 30,
       "score": 83.3
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 32,
       "total": 40,
       "score": 80.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T083339_qwen3.5-4b--unsloth--q8_0",
   "quant_id": "qwen3.5-4b--unsloth--q8_0",
   "started_at": "2026-10-21T08:33:39Z",
   "finished_at": "2026-10-21T08:39:51Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q8_0.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q8_0.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-Q8_0.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.001939,
     "max": 6.198591,
     "p99": 0.010919,
     "median": 0.00055,
     "same_top_pct": 98.31,
     "ppl": 8.908604,
     "ppl_ref": 8.863479,
     "delta_p_rms": 1.402
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 8442.45,
     "pp_stddev": 422.37,
     "tg_tps": 113.95,
     "tg_stddev": 0.14
    },
    "toolcall": {
     "score": 86.5,
     "passed": 173,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 55,
       "total": 60,
       "score": 91.7
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 24,
       "total": 30,
       "score": 80.0
      },
      "parallel_multiple": {
       "passed": 25,
       "total": 30,
       "score": 83.3
      },
      "irrelevance": {
       "passed": 34,
       "total": 40,
       "score": 85.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T084831_qwen3.5-4b--unsloth--ud-q2_k_xl",
   "quant_id": "qwen3.5-4b--unsloth--ud-q2_k_xl",
   "started_at": "2026-10-21T08:48:31Z",
   "finished_at": "2026-10-21T08:54:27Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q2_K_XL.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q2_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q2_K_XL.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.242534,
     "max": 15.634112,
     "p99": 2.036013,
     "median": 0.133113,
     "same_top_pct": 78.62,
     "ppl": 10.553306,
     "ppl_ref": 8.863479,
     "delta_p_rms": 14.337
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 6685.49,
     "pp_stddev": 248.62,
     "tg_tps": 186.52,
     "tg_stddev": 0.62
    },
    "toolcall": {
     "score": 76.0,
     "passed": 152,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 41,
       "total": 60,
       "score": 68.3
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 19,
       "total": 30,
       "score": 63.3
      },
      "parallel_multiple": {
       "passed": 19,
       "total": 30,
       "score": 63.3
      },
      "irrelevance": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T082723_qwen3.5-4b--unsloth--ud-q4_k_xl",
   "quant_id": "qwen3.5-4b--unsloth--ud-q4_k_xl",
   "started_at": "2026-10-21T08:27:23Z",
   "finished_at": "2026-10-21T08:33:39Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.2",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q4_K_XL.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-4b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q4_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q4_K_XL.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-4B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 8.8938,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true
   },
   "results": {
    "kld": {
     "mean": 0.021993,
     "max": 11.916233,
     "p99": 0.190579,
     "median": 0.00955,
     "same_top_pct": 93.302,
     "ppl": 9.063068,
     "ppl_ref": 8.863479,
     "delta_p_rms": 3.962
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 7928.22,
     "pp_stddev": 359.55,
     "tg_tps": 150.61,
     "tg_stddev": 1.82
    },
    "toolcall": {
     "score": 85.0,
     "passed": 170,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 53,
       "total": 60,
       "score": 88.3
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 24,
       "total": 30,
       "score": 80.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 32,
       "total": 40,
       "score": 80.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T063650_qwen3.5-9b--bartowski--iq3_m",
   "quant_id": "qwen3.5-9b--bartowski--iq3_m",
   "started_at": "2026-10-21T06:36:50Z",
   "finished_at": "2026-10-21T06:42:28Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ3_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ3_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ3_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.089902,
     "max": 13.876218,
     "p99": 0.991024,
     "median": 0.038498,
     "same_top_pct": 88.012,
     "ppl": 7.868281,
     "ppl_ref": 7.22977,
     "delta_p_rms": 9.425
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4902.64,
     "pp_stddev": 142.38,
     "tg_tps": 111.95,
     "tg_stddev": 0.15
    },
    "toolcall": {
     "score": 88.0,
     "passed": 176,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 55,
       "total": 60,
       "score": 91.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 27,
       "total": 30,
       "score": 90.0
      },
      "parallel_multiple": {
       "passed": 24,
       "total": 30,
       "score": 80.0
      },
      "irrelevance": {
       "passed": 34,
       "total": 40,
       "score": 85.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T060058_qwen3.5-9b--bartowski--iq4_xs",
   "quant_id": "qwen3.5-9b--bartowski--iq4_xs",
   "started_at": "2026-10-21T06:00:58Z",
   "finished_at": "2026-10-21T06:07:16Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ4_XS.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ4_XS.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-IQ4_XS.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.033208,
     "max": 12.028238,
     "p99": 0.345229,
     "median": 0.011505,
     "same_top_pct": 92.922,
     "ppl": 7.34303,
     "ppl_ref": 7.22977,
     "delta_p_rms": 5.625
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 5357.86,
     "pp_stddev": 166.14,
     "tg_tps": 110.62,
     "tg_stddev": 0.58
    },
    "toolcall": {
     "score": 89.5,
     "passed": 179,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 29,
       "total": 30,
       "score": 96.7
      },
      "parallel_multiple": {
       "passed": 25,
       "total": 30,
       "score": 83.3
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T054144_qwen3.5-9b--bartowski--q4_k_m",
   "quant_id": "qwen3.5-9b--bartowski--q4_k_m",
   "started_at": "2026-10-21T05:41:44Z",
   "finished_at": "2026-10-21T05:48:26Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.021891,
     "max": 10.112214,
     "p99": 0.232341,
     "median": 0.007779,
     "same_top_pct": 94.012,
     "ppl": 7.368461,
     "ppl_ref": 7.22977,
     "delta_p_rms": 4.585
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4651.62,
     "pp_stddev": 134.82,
     "tg_tps": 97.84,
     "tg_stddev": 0.22
    },
    "toolcall": {
     "score": 92.0,
     "passed": 184,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 59,
       "total": 60,
       "score": 98.3
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 32,
       "total": 40,
       "score": 80.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T062938_qwen3.5-9b--bartowski--q5_k_m",
   "quant_id": "qwen3.5-9b--bartowski--q5_k_m",
   "started_at": "2026-10-21T06:29:38Z",
   "finished_at": "2026-10-21T06:36:50Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q5_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q5_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-9B-GGUF\\snapshots\\182be2fd6c7bc44887d88a91cb03ff009cc9f549\\Qwen_Qwen3.5-9B-Q5_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.00933,
     "max": 8.037339,
     "p99": 0.081957,
     "median": 0.003042,
     "same_top_pct": 96.38,
     "ppl": 7.297187,
     "ppl_ref": 7.22977,
     "delta_p_rms": 2.796
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4686.79,
     "pp_stddev": 138.32,
     "tg_tps": 87.52,
     "tg_stddev": 0.28
    },
    "toolcall": {
     "score": 92.5,
     "passed": 185,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 38,
       "total": 40,
       "score": 95.0
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 27,
       "total": 30,
       "score": 90.0
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T060717_qwen3.5-9b--mradermacher--i1-iq4_xs",
   "quant_id": "qwen3.5-9b--mradermacher--i1-iq4_xs",
   "started_at": "2026-10-21T06:07:17Z",
   "finished_at": "2026-10-21T06:14:32Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-IQ4_XS.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-IQ4_XS.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-IQ4_XS.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.037898,
     "max": 13.782431,
     "p99": 0.403085,
     "median": 0.012617,
     "same_top_pct": 92.443,
     "ppl": 7.40089,
     "ppl_ref": 7.22977,
     "delta_p_rms": 6.117
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 5327.21,
     "pp_stddev": 164.4,
     "tg_tps": 113.61,
     "tg_stddev": 0.13
    },
    "toolcall": {
     "score": 89.5,
     "passed": 179,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 28,
       "total": 30,
       "score": 93.3
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 32,
       "total": 40,
       "score": 80.0
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T054826_qwen3.5-9b--mradermacher--i1-q4_k_m",
   "quant_id": "qwen3.5-9b--mradermacher--i1-q4_k_m",
   "started_at": "2026-10-21T05:48:26Z",
   "finished_at": "2026-10-21T05:54:15Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.048088,
     "max": 20.928278,
     "p99": 0.522357,
     "median": 0.013989,
     "same_top_pct": 92.016,
     "ppl": 7.354393,
     "ppl_ref": 7.22977,
     "delta_p_rms": 6.414
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4706.93,
     "pp_stddev": 126.9,
     "tg_tps": 101.4,
     "tg_stddev": 1.15
    },
    "toolcall": {
     "score": 92.0,
     "passed": 184,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T053418_qwen3.5-9b--unsloth--q4_k_m",
   "quant_id": "qwen3.5-9b--unsloth--q4_k_m",
   "started_at": "2026-10-21T05:34:18Z",
   "finished_at": "2026-10-21T05:41:44Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q4_K_M.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q4_K_M.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    },
    "toolcall_parallel_tool_calls": true,
    "toolcall_rerun_at": "2026-10-21T08:12:34Z",
    "toolcall_harness_version": "0.1.2"
   },
   "results": {
    "kld": {
     "mean": 0.046212,
     "max": 20.196701,
     "p99": 0.475209,
     "median": 0.013264,
     "same_top_pct": 92.443,
     "ppl": 7.364546,
     "ppl_ref": 7.22977,
     "delta_p_rms": 6.225
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4699.97,
     "pp_stddev": 128.7,
     "tg_tps": 100.97,
     "tg_stddev": 0.18
    },
    "toolcall": {
     "score": 91.5,
     "passed": 183,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl",
    "toolcall_v0_1_1": "toolcall.v0.1.1.jsonl"
   },
   "example": false,
   "notes": "tool-call stage re-run 2026-10-21T08:12:34Z with parallel_tool_calls=true (first pass used llama-server's default, which allows at most one call per turn; old score 91.5%)"
  },
  {
   "id": "20261021T062206_qwen3.5-9b--unsloth--q6_k",
   "quant_id": "qwen3.5-9b--unsloth--q6_k",
   "started_at": "2026-10-21T06:22:06Z",
   "finished_at": "2026-10-21T06:29:37Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q6_K.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q6_K.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q6_K.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.004256,
     "max": 8.825918,
     "p99": 0.032281,
     "median": 0.00139,
     "same_top_pct": 97.482,
     "ppl": 7.269522,
     "ppl_ref": 7.22977,
     "delta_p_rms": 1.802
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4010.17,
     "pp_stddev": 93.63,
     "tg_tps": 81.95,
     "tg_stddev": 0.17
    },
    "toolcall": {
     "score": 92.5,
     "passed": 185,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 27,
       "total": 30,
       "score": 90.0
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T061432_qwen3.5-9b--unsloth--q8_0",
   "quant_id": "qwen3.5-9b--unsloth--q8_0",
   "started_at": "2026-10-21T06:14:32Z",
   "finished_at": "2026-10-21T06:22:05Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q8_0.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q8_0.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q8_0.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.002487,
     "max": 5.891551,
     "p99": 0.014146,
     "median": 0.000527,
     "same_top_pct": 98.435,
     "ppl": 7.273867,
     "ppl_ref": 7.22977,
     "delta_p_rms": 1.265
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4985.22,
     "pp_stddev": 147.08,
     "tg_tps": 66.43,
     "tg_stddev": 0.19
    },
    "toolcall": {
     "score": 91.5,
     "passed": 183,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 57,
       "total": 60,
       "score": 95.0
      },
      "multiple": {
       "passed": 37,
       "total": 40,
       "score": 92.5
      },
      "parallel": {
       "passed": 30,
       "total": 30,
       "score": 100.0
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T064229_qwen3.5-9b--unsloth--ud-q2_k_xl",
   "quant_id": "qwen3.5-9b--unsloth--ud-q2_k_xl",
   "started_at": "2026-10-21T06:42:29Z",
   "finished_at": "2026-10-21T06:49:58Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q2_K_XL.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q2_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q2_K_XL.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.207126,
     "max": 13.780758,
     "p99": 2.177594,
     "median": 0.099432,
     "same_top_pct": 81.412,
     "ppl": 8.519052,
     "ppl_ref": 7.22977,
     "delta_p_rms": 14.591
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 3931.94,
     "pp_stddev": 91.79,
     "tg_tps": 132.98,
     "tg_stddev": 0.32
    },
    "toolcall": {
     "score": 82.0,
     "passed": 164,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 36,
       "total": 40,
       "score": 90.0
      },
      "parallel": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "parallel_multiple": {
       "passed": 17,
       "total": 30,
       "score": 56.7
      },
      "irrelevance": {
       "passed": 27,
       "total": 40,
       "score": 67.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  },
  {
   "id": "20261021T055415_qwen3.5-9b--unsloth--ud-q4_k_xl",
   "quant_id": "qwen3.5-9b--unsloth--ud-q4_k_xl",
   "started_at": "2026-10-21T05:54:15Z",
   "finished_at": "2026-10-21T06:00:58Z",
   "status": "partial",
   "machine": {
    "name": "Ghost",
    "gpu": "NVIDIA GeForce RTX 5070",
    "vram_gb": 12,
    "ram_gb": 64,
    "cpu": "AMD Ryzen 7 5800X",
    "os": "Windows 11",
    "driver": "616.92",
    "cuda": "13.4"
   },
   "software": {
    "llama_cpp_build": "b11770 (39da130fc)",
    "lm_eval_version": "0.4.13",
    "harness_version": "0.1.1",
    "python": "3.12.10"
   },
   "placement": "full_gpu",
   "settings": {
    "commands": [
     "C:\\quantdiff\\llama.cpp\\bin\\llama-perplexity.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q4_K_XL.gguf --kl-divergence-base C:\\quantdiff\\work\\reference\\qwen3.5-9b\\quantdiff-kld-v1.kld --kl-divergence -c 512 -b 512 -ngl 999",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-bench.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q4_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json",
     "C:\\quantdiff\\llama.cpp\\bin\\llama-server.exe -m C:\\quantdiff\\hf-cache\\models--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-UD-Q4_K_XL.gguf --host 127.0.0.1 --port 8089 -c 4096 -np 1 -ngl 999 -fa on --jinja --seed 1234 --no-webui -a quant"
    ],
    "ctx": 4096,
    "batch": 512,
    "threads": 8,
    "flash_attn": true,
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "seed": 1234,
    "temperature": 0.0,
    "kld_corpus": "quantdiff-kld-v1",
    "kld_corpus_sha256": "ff19568db86d014569a44b6da4c5830fbcd26de59f57c9ece87b45c34d9c5782",
    "kld_chunks": 100,
    "kld_ctx": 512,
    "kld_reference": "Qwen/Qwen3.5-9B -> convert_hf_to_gguf.py --outtype bf16 (cached)",
    "kld_ref_ppl": 7.257,
    "ngl": 999,
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "server_ctx": 4096,
    "server_ngl": 999,
    "toolcall_set": "quantdiff-bfcl-v1",
    "toolcall_set_sha256": "4d0e1667163c96a109965c58242673528c62318322450d9b218a730c09965020",
    "toolcall_chat_template_kwargs": {
     "enable_thinking": false
    }
   },
   "results": {
    "kld": {
     "mean": 0.020581,
     "max": 8.168179,
     "p99": 0.225234,
     "median": 0.007052,
     "same_top_pct": 94.443,
     "ppl": 7.34036,
     "ppl_ref": 7.22977,
     "delta_p_rms": 4.494
    },
    "speed": {
     "pp_tokens": 512,
     "tg_tokens": 128,
     "pp_tps": 4719.22,
     "pp_stddev": 271.63,
     "tg_tps": 97.27,
     "tg_stddev": 0.16
    },
    "toolcall": {
     "score": 90.5,
     "passed": 181,
     "total": 200,
     "request_errors": 0,
     "by_category": {
      "simple_python": {
       "passed": 58,
       "total": 60,
       "score": 96.7
      },
      "multiple": {
       "passed": 35,
       "total": 40,
       "score": 87.5
      },
      "parallel": {
       "passed": 29,
       "total": 30,
       "score": 96.7
      },
      "parallel_multiple": {
       "passed": 26,
       "total": 30,
       "score": 86.7
      },
      "irrelevance": {
       "passed": 33,
       "total": 40,
       "score": 82.5
      }
     }
    }
   },
   "logs": {
    "kld": "kld.log",
    "llama_bench": "llama_bench.log",
    "llama_server": "llama_server.log",
    "toolcall": "toolcall.jsonl"
   },
   "example": false,
   "notes": ""
  }
 ]
}