Nex-N2.5-mini-ROCmFP4-GGUF / recipe /results_summary.json
kingjones777's picture
recipe: BPW footnote inputs (GGUF weight counts) + card scripts
5dba7e1 verified
Raw
History Blame
59.3 kB
{
"layout": "no MTP (checkpoint declares mtp_num_hidden_layers=1 but ships no mtp.* tensors)",
"model": "Nex-N2.5-mini",
"model_repo_std": "kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF",
"model_repo_imat": "kingjones777/Nex-N2.5-mini-ROCmFP4-imatrix-GGUF",
"source": {
"repo": "nex-agi/Nex-N2.5-mini",
"revision": "87420286149d9cce9bd46cd335ef9bda33c37c1b",
"license": "apache-2.0",
"gated": false,
"params": 35107181936
},
"arch": {
"name": "qwen3_5_moe",
"llama_cpp": "qwen35moe",
"layers": 40,
"linear_attn_layers": 30,
"full_attn_layers": 10,
"full_attention_interval": 4,
"hidden": 2048,
"num_experts": 256,
"num_experts_per_tok": 8,
"expert_width": 512,
"shared_expert_width": 512,
"attn_heads": 16,
"kv_heads": 2,
"head_dim": 256,
"linear_key_heads": 16,
"linear_value_heads": 32,
"linear_head_dim": 128,
"vocab": 248320,
"max_position_embeddings": 262144,
"tie_word_embeddings": false,
"vision_layers": 27,
"vision_width": 1152,
"hf_tensors": 1026,
"mtp_num_hidden_layers_declared": 1
},
"bf16": {
"ppl": 6.2303,
"ppl_err": 0.07538,
"ppl_paired": 6.228979,
"ppl_paired_err": 0.075322,
"chunks": 40,
"n_ctx": 2048,
"scored_tokens": 40920,
"size_bytes": 69376636928,
"elements": 34660610688,
"bpw_logged": "16.01",
"file": "Nex-N2.5-mini-BF16.gguf",
"readback": "PASS",
"arch": "qwen35moe",
"ftype": 32,
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "BF16",
"token_embd": "BF16",
"mtp_named_tensors": 0,
"nextn_named_tensors": 0
},
"binary": {
"repo": "https://github.com/charlie12345/ROCmFPX",
"commit": "d3ca537",
"dir": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"sha256": {
"llama-server": "d077b98f3f04df4b482a5826c8922e39cda3aa8d2e2b2fbe1870f36d245cebe7",
"llama-quantize": "1559189968a30781f837e4d25ca3cdf26c8eb0351023675406535eddcbe0db13",
"llama-imatrix": "bfdce5c21a01bdc8043d69335b74a7b68b8335ae989991a1e10e21d5faeaeb8e",
"llama-perplexity": "c2175925bb0c61df89ff787845d8e03a35403f684be52719ac2987b8ca636a53"
}
},
"vision_probe": {
"width": 448,
"height": 448,
"bytes": 1945,
"sha256": "53924b4715db9757d5b70614dd0e560b8f7de686ff3df9792721d157e2594f5e"
},
"imatrix": {
"chunks": 129,
"n_ctx": 512,
"calibration": "bartowski calibration_datav3.txt",
"file": "Nex-N2.5-mini.imatrix",
"sha256": "7e5afffc822c64a7b43a1223d0094fb60895a521f6c89b912409d25091bd17f8",
"size_bytes": 192223936,
"device": "CPU"
},
"hub_check": {
"checked_at_utc": "2026-09-16T21:27:05Z",
"queries": [
"Nex-N2.5-mini",
"Nex-N2.5",
"nex-n2.5-mini gguf",
"Nex N2.5 mini"
],
"repos": [
{
"repo": "6block/Nex-N2.5-Pro-GGUF",
"gguf_files": 58,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-Pro-IQ2_XS-00001-of-00019.gguf",
"Nex-N2.5-Pro-IQ2_XS-00002-of-00019.gguf",
"Nex-N2.5-Pro-IQ2_XS-00003-of-00019.gguf"
]
},
{
"repo": "Arahide/Nex-N2.5-mini-INT4-W4A16",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "DevQuasar/nex-agi.Nex-N2.5-Max-GGUF",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "DevQuasar/nex-agi.Nex-N2.5-Pro-GGUF",
"gguf_files": 66,
"rocm_like_files": [],
"gguf_sample": [
"Q2_K/nex-agi.Nex-N2.5-Pro.f16.gguf.Q2_K.gguf-00001-of-00011.gguf",
"Q2_K/nex-agi.Nex-N2.5-Pro.f16.gguf.Q2_K.gguf-00002-of-00011.gguf",
"Q2_K/nex-agi.Nex-N2.5-Pro.f16.gguf.Q2_K.gguf-00003-of-00011.gguf"
]
},
{
"repo": "DevQuasar/nex-agi.Nex-N2.5-mini-GGUF",
"gguf_files": 7,
"rocm_like_files": [],
"gguf_sample": [
"Q2_K/nex-agi.Nex-N2.5-mini.f16.gguf.Q2_K.gguf",
"Q3_K_M/nex-agi.Nex-N2.5-mini.f16.gguf.Q3_K_M.gguf",
"Q4_K_M/nex-agi.Nex-N2.5-mini.f16.gguf.Q4_K_M.gguf"
]
},
{
"repo": "IsValorum/Nex-N2.5-mini-APEX-I-MiniPlus-GGUF",
"gguf_files": 2,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini.APEX-I-MiniPlus.gguf",
"mmproj-nex-agi_Nex-N2.5-mini-Q8_0.gguf"
]
},
{
"repo": "Kagandi/Nex-N2.5-mini-mlx-4Bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "MrFuzzihead/Nex-N2.5-mini-APEX-GGUF",
"gguf_files": 12,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-APEX-Balanced.gguf",
"Nex-N2.5-mini-APEX-Compact.gguf",
"Nex-N2.5-mini-APEX-I-Balanced.gguf"
]
},
{
"repo": "NandoG-AI/Nex-N2.5-mini-GGUF",
"gguf_files": 1,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-Q4_K_M.gguf"
]
},
{
"repo": "ProCreations/Nex-N2.5-mini-NVFP4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "SAIFIINDUSTRIES/Nex-N2.5-mini",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Tech2wild/Nex-N2.5-mini-Int4-Int8Mix",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-4bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-6bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-8bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-oQ2",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-oQ3",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-oQ4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-oQ6",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "Vontra/Nex-N2.5-mini-MLX-oQ8",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "abenzerps/Nex-N2.5-mini-GGUF",
"gguf_files": 18,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-IQ1_M.gguf",
"Nex-N2.5-mini-IQ1_S.gguf",
"Nex-N2.5-mini-IQ2_M.gguf"
]
},
{
"repo": "abenzerps/Nex-N2.5-mini-MLX-4bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "abenzerps/Nex-N2.5-mini-MLX-6bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "abenzerps/Nex-N2.5-mini-MLX-8bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "auryn-macmillan/Nex-N2.5-mini-exl3-5bpw",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "bartowski/Nex-N2.5-Pro-GGUF",
"gguf_files": 144,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-Pro-IQ1_M/Nex-N2.5-Pro-IQ1_M-00001-of-00003.gguf",
"Nex-N2.5-Pro-IQ1_M/Nex-N2.5-Pro-IQ1_M-00002-of-00003.gguf",
"Nex-N2.5-Pro-IQ1_M/Nex-N2.5-Pro-IQ1_M-00003-of-00003.gguf"
]
},
{
"repo": "bartowski/nex-agi_Nex-N2.5-mini-GGUF",
"gguf_files": 28,
"rocm_like_files": [],
"gguf_sample": [
"mmproj-nex-agi_Nex-N2.5-mini-bf16.gguf",
"mmproj-nex-agi_Nex-N2.5-mini-f16.gguf",
"nex-agi_Nex-N2.5-mini-IQ2_M.gguf"
]
},
{
"repo": "cbert33/Nex-N2.5-mini-FP8-Calibrated",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "ghazni101/Nex-N2.5-mini-MQ4R",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "julianmb/Nex-N2.5-mini-ROCmFP4-GGUF",
"gguf_files": 2,
"rocm_like_files": [
"Nex-N2.5-mini-ROCmFP4-STRIX_LEAN.gguf"
],
"gguf_sample": [
"Nex-N2.5-mini-ROCmFP4-STRIX_LEAN.gguf",
"mmproj-Nex-N2.5-mini.gguf"
]
},
{
"repo": "karmx/Nex-N2.5-mini-Mixed-Q2Q3-128K-GGUF",
"gguf_files": 2,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-mixed-Q2Q3-vision-trial.gguf",
"mmproj-Nex-N2.5-mini-f16.gguf"
]
},
{
"repo": "liskasYR/Nex-N2.5-Max",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "mlx-community/Nex-N2.5-mini-OptiQ-4bit",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "mlx-community/Nex-N2.5-mini-oQ4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "mradermacher/Nex-N2.5-mini-GGUF",
"gguf_files": 13,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini.IQ4_XS.gguf",
"Nex-N2.5-mini.Q2_K.gguf",
"Nex-N2.5-mini.Q3_K_L.gguf"
]
},
{
"repo": "mradermacher/Nex-N2.5-mini-i1-GGUF",
"gguf_files": 24,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini.i1-IQ1_M.gguf",
"Nex-N2.5-mini.i1-IQ1_S.gguf",
"Nex-N2.5-mini.i1-IQ2_M.gguf"
]
},
{
"repo": "nex-agi/Nex-N2.5-Max",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "nex-agi/Nex-N2.5-mini",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "ngquocvinh/Nex-N2.5-mini-GGUF",
"gguf_files": 16,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-IQ1_M.gguf",
"Nex-N2.5-mini-IQ2_XS.gguf",
"Nex-N2.5-mini-IQ3_M.gguf"
]
},
{
"repo": "orcarouter/Nex-N2.5-mini-Uncensored",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "orcarouter/Nex-N2.5-mini-Uncensored-FP8",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "orcarouter/Nex-N2.5-mini-Uncensored-GGUF",
"gguf_files": 6,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-Uncensored-IQ4_XS.gguf",
"Nex-N2.5-mini-Uncensored-Q2_K.gguf",
"Nex-N2.5-mini-Uncensored-Q3_K_M.gguf"
]
},
{
"repo": "orcarouter/Nex-N2.5-mini-Uncensored-MLX",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "orcarouter/Nex-N2.5-mini-Uncensored-NVFP4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "primitive-ai/Nex-N2.5-mini-FP8",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "primitive-ai/Nex-N2.5-mini-NVFP4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "primitive-ai/Nex-N2.5-mini-mixed-NVFP4-FP8",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "qtum/Nex-N2.5-Pro-GGUF",
"gguf_files": 58,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-Pro-IQ2_XS-00001-of-00019.gguf",
"Nex-N2.5-Pro-IQ2_XS-00002-of-00019.gguf",
"Nex-N2.5-Pro-IQ2_XS-00003-of-00019.gguf"
]
},
{
"repo": "quant-mind/Nex-N2.5-mini-W4A16-AutoRound",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "sakamakismile/Nex-N2.5-Max-GGUF",
"gguf_files": 39,
"rocm_like_files": [],
"gguf_sample": [
"Q3_K_M/Nex-N2.5-Max-Q3_K_M-00001-of-00018.gguf",
"Q3_K_M/Nex-N2.5-Max-Q3_K_M-00002-of-00018.gguf",
"Q3_K_M/Nex-N2.5-Max-Q3_K_M-00003-of-00018.gguf"
]
},
{
"repo": "sigmanih/nex-agi-Nex-N2.5-mini-GGUF-Q4_K_S",
"gguf_files": 1,
"rocm_like_files": [],
"gguf_sample": [
"nex-agi--Nex-N2.5-mini.Q4_K_S.gguf"
]
},
{
"repo": "sojufx/Nex-N2.5-mini-NVFP4",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "suzu89/Nex-N2.5-mini-oQ5",
"gguf_files": 0,
"rocm_like_files": [],
"gguf_sample": []
},
{
"repo": "vikrant1123/Nex-N2.5-mini-Uncensored-APEX-GGUF",
"gguf_files": 5,
"rocm_like_files": [],
"gguf_sample": [
"Nex-N2.5-mini-Uncensored-APEX-I-Balanced.gguf",
"Nex-N2.5-mini-Uncensored-APEX-I-Compact.gguf",
"Nex-N2.5-mini-Uncensored-APEX-I-Quality.gguf"
]
}
],
"gguf_repos": 19,
"rocm_builds_found": 1,
"header_checks": [
{
"repo": "julianmb/Nex-N2.5-mini-ROCmFP4-GGUF",
"file": "Nex-N2.5-mini-ROCmFP4-STRIX_LEAN.gguf",
"file_type": 106,
"tensors": 733,
"block_count": 40,
"output_weight_type_id": 101,
"token_embd_type_id": 13,
"imatrix_keys": [],
"read_via": "HTTP range read of the GGUF header",
"output_weight_type": "Q4_0_ROCMFP4_FAST",
"token_embd_type": "Q5_K"
}
]
},
"repeat": {
"step": "N5",
"device": "ROCm0",
"rows": 40,
"result": "MATCH"
},
"repeat_vk": {
"step": "N5v",
"device": "Vulkan0",
"rows": 40,
"result": "MATCH"
},
"reference": {
"step": "N1c",
"cpu_chunk1": 5.6964,
"vulkan0_chunk1": 5.6953,
"hip_rocm0_chunk1": 139.1466,
"hip_rocm0_faoff_chunk1": 136.6879,
"hip_rocm0_final": 105.9103,
"result": "PASS"
},
"reference_device": "CPU",
"tiers": {
"q106": {
"file": "Nex-N2.5-mini-Q4_0-ROCmFP4-STRIX_LEAN.gguf",
"local_file": "Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"dir": "out",
"ftype": 106,
"size_bytes": 18744342208,
"quant_mib": 17865.52,
"bpw": 4.32,
"quant_seconds": 394.87116,
"imatrix_entries": null,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q5_K",
"quality_measured": true,
"ppl": 6.473964,
"ppl_err": 0.079799,
"ppl_ratio": 1.03933,
"kld_mean": 0.104436,
"kld_err": 0.001413,
"kld_p99": 0.926687,
"kld_median": 0.047702,
"same_top_p": 86.659,
"rms_dp": 9.028,
"vk": {
"ppl": 6.474466,
"ppl_err": 0.079947,
"ppl_ratio": 1.03941,
"kld_mean": 0.104408,
"kld_err": 0.001395,
"kld_p99": 0.899358,
"kld_median": 0.048009,
"same_top_p": 86.794,
"rms_dp": 9.03
}
},
"q102": {
"file": "Nex-N2.5-mini-Q4_0-ROCmFP4-COHERENT.gguf",
"local_file": "Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf",
"dir": "out",
"ftype": 102,
"size_bytes": 19846166208,
"quant_mib": 18916.3,
"bpw": 4.58,
"quant_seconds": 443.71327,
"imatrix_entries": null,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q6_K",
"quality_measured": true,
"ppl": 6.561727,
"ppl_err": 0.08124,
"ppl_ratio": 1.053419,
"kld_mean": 0.097071,
"kld_err": 0.001342,
"kld_p99": 0.874485,
"kld_median": 0.043524,
"same_top_p": 87.287,
"rms_dp": 8.8,
"vk": {
"ppl": 6.592123,
"ppl_err": 0.081874,
"ppl_ratio": 1.058299,
"kld_mean": 0.097165,
"kld_err": 0.001337,
"kld_p99": 0.858566,
"kld_median": 0.043968,
"same_top_p": 87.265,
"rms_dp": 8.853
}
},
"q103": {
"file": "Nex-N2.5-mini-Q4_0-ROCmFP4-FAST.gguf",
"local_file": "Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf",
"dir": "out",
"ftype": 103,
"size_bytes": 18648495808,
"quant_mib": 17774.11,
"bpw": 4.3,
"quant_seconds": 392.98796999999996,
"imatrix_entries": null,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q4_0_ROCMFP4_FAST",
"quality_measured": true,
"ppl": 6.549829,
"ppl_err": 0.080919,
"ppl_ratio": 1.051509,
"kld_mean": 0.108779,
"kld_err": 0.00142,
"kld_p99": 0.956811,
"kld_median": 0.04973,
"same_top_p": 86.405,
"rms_dp": 9.108,
"vk": {
"ppl": 6.551305,
"ppl_err": 0.081055,
"ppl_ratio": 1.051746,
"kld_mean": 0.108756,
"kld_err": 0.001422,
"kld_p99": 0.930024,
"kld_median": 0.050197,
"same_top_p": 86.356,
"rms_dp": 9.101
}
},
"q106i": {
"file": "Nex-N2.5-mini-imatrix-Q4_0-ROCmFP4-STRIX_LEAN.gguf",
"local_file": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"dir": "out-imat",
"ftype": 106,
"size_bytes": 18744342464,
"quant_mib": 17865.52,
"bpw": 4.32,
"quant_seconds": 306.08412,
"imatrix_entries": 510,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q5_K",
"quality_measured": true,
"ppl": 6.353616,
"ppl_err": 0.077001,
"ppl_ratio": 1.020009,
"kld_mean": 0.085158,
"kld_err": 0.001309,
"kld_p99": 0.737186,
"kld_median": 0.038436,
"same_top_p": 87.849,
"rms_dp": 8.281,
"vk": {
"ppl": 6.364168,
"ppl_err": 0.077334,
"ppl_ratio": 1.021703,
"kld_mean": 0.083563,
"kld_err": 0.001206,
"kld_p99": 0.713036,
"kld_median": 0.038,
"same_top_p": 87.903,
"rms_dp": 8.201
},
"same_tensor_types_as_standard": true,
"same_tensor_names_types_bytes_as_standard": true,
"file_size_delta_bytes": 256,
"differs_from_standard": true,
"header_keys_only_in_imatrix": [
"quantize.imatrix.chunks_count",
"quantize.imatrix.dataset",
"quantize.imatrix.entries_count",
"quantize.imatrix.file"
],
"header_keys_only_in_standard": []
},
"q102i": {
"file": "Nex-N2.5-mini-imatrix-Q4_0-ROCmFP4-COHERENT.gguf",
"local_file": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf",
"dir": "out-imat",
"ftype": 102,
"size_bytes": 19846166464,
"quant_mib": 18916.3,
"bpw": 4.58,
"quant_seconds": 409.08034000000004,
"imatrix_entries": 510,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q6_K",
"quality_measured": true,
"ppl": 6.36005,
"ppl_err": 0.077045,
"ppl_ratio": 1.021042,
"kld_mean": 0.076947,
"kld_err": 0.001181,
"kld_p99": 0.685225,
"kld_median": 0.034241,
"same_top_p": 88.463,
"rms_dp": 7.909,
"vk": {
"ppl": 6.36321,
"ppl_err": 0.077202,
"ppl_ratio": 1.021549,
"kld_mean": 0.076844,
"kld_err": 0.001332,
"kld_p99": 0.669352,
"kld_median": 0.033844,
"same_top_p": 88.556,
"rms_dp": 7.783
},
"same_tensor_types_as_standard": true,
"same_tensor_names_types_bytes_as_standard": true,
"file_size_delta_bytes": 256,
"differs_from_standard": true,
"header_keys_only_in_imatrix": [
"quantize.imatrix.chunks_count",
"quantize.imatrix.dataset",
"quantize.imatrix.entries_count",
"quantize.imatrix.file"
],
"header_keys_only_in_standard": []
},
"q103i": {
"file": "Nex-N2.5-mini-imatrix-Q4_0-ROCmFP4-FAST.gguf",
"local_file": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf",
"dir": "out-imat",
"ftype": 103,
"size_bytes": 18648496064,
"quant_mib": 17774.11,
"bpw": 4.3,
"quant_seconds": 328.50694,
"imatrix_entries": 510,
"readback": "PASS",
"arch": "qwen35moe",
"tensors": 733,
"nextn_tensors": 0,
"output_weight": "Q6_K",
"token_embd": "Q4_0_ROCMFP4_FAST",
"quality_measured": true,
"ppl": 6.368075,
"ppl_err": 0.077317,
"ppl_ratio": 1.02233,
"kld_mean": 0.088974,
"kld_err": 0.001318,
"kld_p99": 0.784672,
"kld_median": 0.039738,
"same_top_p": 87.454,
"rms_dp": 8.37,
"vk": {
"ppl": 6.392929,
"ppl_err": 0.077851,
"ppl_ratio": 1.026321,
"kld_mean": 0.08915,
"kld_err": 0.001266,
"kld_p99": 0.784911,
"kld_median": 0.039583,
"same_top_p": 87.424,
"rms_dp": 8.509
},
"same_tensor_types_as_standard": true,
"same_tensor_names_types_bytes_as_standard": true,
"file_size_delta_bytes": 256,
"differs_from_standard": true,
"header_keys_only_in_imatrix": [
"quantize.imatrix.chunks_count",
"quantize.imatrix.dataset",
"quantize.imatrix.entries_count",
"quantize.imatrix.file"
],
"header_keys_only_in_standard": []
}
},
"aux": {
"mmproj-Nex-N2.5-mini-BF16.gguf": 902821920,
"chat_template_enable_thinking.jinja": 7895
},
"mmproj": {
"file": "mmproj-Nex-N2.5-mini-BF16.gguf",
"size_bytes": 902821920,
"elements": 446571248,
"readback": "PASS",
"arch": "clip",
"ftype": 32,
"tensors": 334
},
"bench": [
{
"label": "n-q106-rocm",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 63.94,
"tg_min": 63.56,
"tg_max": 63.96,
"pp_median": 1158.2,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7098,
"prompt_n_max": 7100
},
{
"label": "n-q106-vk",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 4.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 68.16,
"tg_min": 68.1,
"tg_max": 68.21,
"pp_median": 1006.0,
"prompt_n": 7096,
"accept": null,
"prompt_n_min": 7096,
"prompt_n_max": 7098
},
{
"label": "n-q102-rocm",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 24.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 61.74,
"tg_min": 61.74,
"tg_max": 61.8,
"pp_median": 1186.7,
"prompt_n": 7100,
"accept": null,
"prompt_n_min": 7096,
"prompt_n_max": 7100
},
{
"label": "n-q102-vk",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 68.0,
"tg_min": 67.88,
"tg_max": 68.09,
"pp_median": 1001.4,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7098,
"prompt_n_max": 7101
},
{
"label": "n-q103-rocm",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 63.31,
"tg_min": 63.14,
"tg_max": 63.38,
"pp_median": 1156.6,
"prompt_n": 7100,
"accept": null,
"prompt_n_min": 7094,
"prompt_n_max": 7100
},
{
"label": "n-q103-vk",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 20.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 68.62,
"tg_min": 68.6,
"tg_max": 68.67,
"pp_median": 998.9,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7098,
"prompt_n_max": 7101
},
{
"label": "n-q106i-rocm",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 63.29,
"tg_min": 63.19,
"tg_max": 63.41,
"pp_median": 1145.3,
"prompt_n": 7100,
"accept": null,
"prompt_n_min": 7097,
"prompt_n_max": 7100
},
{
"label": "n-q106i-vk",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 20.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 67.77,
"tg_min": 67.74,
"tg_max": 67.86,
"pp_median": 995.0,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7098,
"prompt_n_max": 7100
},
{
"label": "n-q102i-rocm",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 24.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 61.56,
"tg_min": 61.53,
"tg_max": 61.63,
"pp_median": 1183.6,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7096,
"prompt_n_max": 7100
},
{
"label": "n-q102i-vk",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 67.81,
"tg_min": 67.77,
"tg_max": 67.85,
"pp_median": 998.7,
"prompt_n": 7097,
"accept": null,
"prompt_n_min": 7096,
"prompt_n_max": 7097
},
{
"label": "n-q103i-rocm",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "code",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 63.03,
"tg_min": 62.55,
"tg_max": 63.11,
"pp_median": 1154.4,
"prompt_n": 7101,
"accept": null,
"prompt_n_min": 7099,
"prompt_n_max": 7101
},
{
"label": "n-q103i-vk",
"model": "Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "code",
"load_s": 20.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out-imat/Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 68.44,
"tg_min": 68.42,
"tg_max": 68.56,
"pp_median": 993.9,
"prompt_n": 7098,
"accept": null,
"prompt_n_min": 7098,
"prompt_n_max": 7102
},
{
"label": "n-q106-rocm-prose",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "ROCm0",
"ctx": 65536,
"workload": "prose",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev ROCm0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 63.06,
"tg_min": 62.94,
"tg_max": 63.09,
"pp_median": 1139.1,
"prompt_n": 7454,
"accept": null,
"prompt_n_min": 7453,
"prompt_n_max": 7455
},
{
"label": "n-q106-vk-prose",
"model": "Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf",
"draft": null,
"nmax": null,
"strict": false,
"bin": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin",
"dev": "Vulkan0",
"ctx": 65536,
"workload": "prose",
"load_s": 22.0,
"cmd": "/opt/llama-rocm/rocmfpx-724/build-hipvk/bin/llama-server -m /mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf -dev Vulkan0 -ngl 999 -fa on -dio --jinja -fit off --parallel 1 -c 65536 -b 2048 -ub 1024 --host 127.0.0.1 --port 18600 --no-webui",
"tg_median": 67.34,
"tg_min": 67.31,
"tg_max": 67.35,
"pp_median": 977.7,
"prompt_n": 7454,
"accept": null,
"prompt_n_min": 7454,
"prompt_n_max": 7457
}
],
"gates": [
{
"label": "n-c3-q106",
"gate": "cachegate3-aligned",
"n": 5,
"L": 7039,
"expect_cache_n": 6011,
"all_reused": true,
"all_aligned": true,
"identical": 5,
"rows": [
{
"variant": 0,
"L": 7039,
"warm_prompt_n": 1028,
"warm_cache_n": 6011,
"cold_prompt_n": 7039,
"cold_cache_n": 0,
"aligned": true,
"warm_draft": [
null,
null
],
"cold_draft": [
null,
null
],
"identical": true,
"first_diff_char": null,
"warm_sha": "013b7b4662ba",
"cold_sha": "013b7b4662ba",
"instr": "Write a function that lists every tensor name in this file. ok ok ok ok",
"warm_instr": "Summarise this file. (warm-up 0) ok ok ok ok",
"warm_text": "```python\ndef list_tensor_names(model: ModelBase) -> list[str]:\n \"\"\"Return every tensor name indexed by the model, in model-tensor order.\"\"\"\n return list(model.model_tensors.keys())\n```\n\nExample usage:\n\n```python\nmodel = SomeModelClass(\n dir_model=Path(\"...\"),\n ftype=gguf.LlamaFileType.GUESSED,\n fname_out=Path(\"model.gguf\"),\n)\n\nfor name in list_tensor_names(model):\n print(name)\n```\n\nIf you want to include tensors that are only produced later by `modify_tensors()` or quantization dequantization, use this version instead:\n\n```python\ndef list_all_tensor_names(model: ModelBase) -> list[str]:\n \"\"\"Return tensor names from indexing plus tensors produced by tensor transforms.\"\"\"\n names = set(model.model_tensors.keys())\n\n for name, gen in model.model",
"cold_text": "```python\ndef list_tensor_names(model: ModelBase) -> list[str]:\n \"\"\"Return every tensor name indexed by the model, in model-tensor order.\"\"\"\n return list(model.model_tensors.keys())\n```\n\nExample usage:\n\n```python\nmodel = SomeModelClass(\n dir_model=Path(\"...\"),\n ftype=gguf.LlamaFileType.GUESSED,\n fname_out=Path(\"model.gguf\"),\n)\n\nfor name in list_tensor_names(model):\n print(name)\n```\n\nIf you want to include tensors that are only produced later by `modify_tensors()` or quantization dequantization, use this version instead:\n\n```python\ndef list_all_tensor_names(model: ModelBase) -> list[str]:\n \"\"\"Return tensor names from indexing plus tensors produced by tensor transforms.\"\"\"\n names = set(model.model_tensors.keys())\n\n for name, gen in model.model",
"warm_prompt_ms": 999.89,
"cold_prompt_ms": 6158.66
},
{
"variant": 1,
"L": 7039,
"warm_prompt_n": 1028,
"warm_cache_n": 6011,
"cold_prompt_n": 7039,
"cold_cache_n": 0,
"aligned": true,
"warm_draft": [
null,
null
],
"cold_draft": [
null,
null
],
"identical": true,
"first_diff_char": null,
"warm_sha": "76498f09c9c8",
"cold_sha": "76498f09c9c8",
"instr": "Write a function that counts the model classes registered in this file. ok ok ok",
"warm_instr": "Summarise this file. (warm-up 1) ok ok ok ok",
"warm_text": "```python\ndef count_model_classes(path: str | os.PathLike[str]) -> int:\n \"\"\"Count classes that inherit from ModelBase in the given Python file.\"\"\"\n with open(path, \"r\", encoding=\"utf-8\") as f:\n tree = ast.parse(f.read(), filename=path)\n\n count = 0\n\n for node in tree.body:\n if not isinstance(node, ast.ClassDef):\n continue\n\n for base in node.bases:\n if isinstance(base, ast.Name) and base.id == \"ModelBase\":\n count += 1\n break\n\n return count\n```\n\nExample usage:\n\n```python\nif __name__ == \"__main__\":\n print(count_model_classes(\"your_file.py\"))\n```\n\nThis counts only direct subclasses of `ModelBase`, not indirect subclasses. If you want to count indirect subclasses too, use this version:",
"cold_text": "```python\ndef count_model_classes(path: str | os.PathLike[str]) -> int:\n \"\"\"Count classes that inherit from ModelBase in the given Python file.\"\"\"\n with open(path, \"r\", encoding=\"utf-8\") as f:\n tree = ast.parse(f.read(), filename=path)\n\n count = 0\n\n for node in tree.body:\n if not isinstance(node, ast.ClassDef):\n continue\n\n for base in node.bases:\n if isinstance(base, ast.Name) and base.id == \"ModelBase\":\n count += 1\n break\n\n return count\n```\n\nExample usage:\n\n```python\nif __name__ == \"__main__\":\n print(count_model_classes(\"your_file.py\"))\n```\n\nThis counts only direct subclasses of `ModelBase`, not indirect subclasses. If you want to count indirect subclasses too, use this version:",
"warm_prompt_ms": 1002.97,
"cold_prompt_ms": 6165.72
},
{
"variant": 2,
"L": 7039,
"warm_prompt_n": 1028,
"warm_cache_n": 6011,
"cold_prompt_n": 7039,
"cold_cache_n": 0,
"aligned": true,
"warm_draft": [
null,
null
],
"cold_draft": [
null,
null
],
"identical": true,
"first_diff_char": null,
"warm_sha": "16577469a2ea",
"cold_sha": "16577469a2ea",
"instr": "Write a function that finds the longest method in this file. ok ok ok ok",
"warm_instr": "Summarise this file. (warm-up 2) ok ok ok ok",
"warm_text": "```python\nimport ast\nfrom pathlib import Path\n\n\ndef find_longest_method(path: str | Path) -> tuple[str, int, ast.FunctionDef | ast.AsyncFunctionDef]:\n \"\"\"\n Return the longest method in a Python file.\n\n Length is measured by the number of AST nodes in the method body.\n \"\"\"\n source_path = Path(path)\n tree = ast.parse(source_path.read_text(encoding=\"utf-8\"), filename=str(source_path))\n\n methods = [\n node\n for node in ast.walk(tree)\n if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))\n ]\n\n if not methods:\n raise ValueError(f\"No methods found in {source_path}\")\n\n longest = max(\n methods,\n key=lambda node: sum(1 for _ in ast.walk(node.body)),\n )\n\n return (\n longest.name",
"cold_text": "```python\nimport ast\nfrom pathlib import Path\n\n\ndef find_longest_method(path: str | Path) -> tuple[str, int, ast.FunctionDef | ast.AsyncFunctionDef]:\n \"\"\"\n Return the longest method in a Python file.\n\n Length is measured by the number of AST nodes in the method body.\n \"\"\"\n source_path = Path(path)\n tree = ast.parse(source_path.read_text(encoding=\"utf-8\"), filename=str(source_path))\n\n methods = [\n node\n for node in ast.walk(tree)\n if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))\n ]\n\n if not methods:\n raise ValueError(f\"No methods found in {source_path}\")\n\n longest = max(\n methods,\n key=lambda node: sum(1 for _ in ast.walk(node.body)),\n )\n\n return (\n longest.name",
"warm_prompt_ms": 1003.48,
"cold_prompt_ms": 6169.04
},
{
"variant": 3,
"L": 7039,
"warm_prompt_n": 1028,
"warm_cache_n": 6011,
"cold_prompt_n": 7039,
"cold_cache_n": 0,
"aligned": true,
"warm_draft": [
null,
null
],
"cold_draft": [
null,
null
],
"identical": true,
"first_diff_char": null,
"warm_sha": "a8545d39788d",
"cold_sha": "a8545d39788d",
"instr": "Write a function that returns every regular expression used in this file. ok ok ok",
"warm_instr": "Summarise this file. (warm-up 3) ok ok ok ok",
"warm_text": "```python\nimport re\n\ndef extract_regular_expressions(source: str) -> list[str]:\n \"\"\"\n Return every regular expression string literal used in the given Python source.\n\n This is a lexical scanner, not a full AST parser, so it handles normal string\n literals but does not evaluate them.\n \"\"\"\n expressions: list[str] = []\n i = 0\n n = len(source)\n\n while i < n:\n if source[i] not in \"'\\\"\":\n i += 1\n continue\n\n quote = source[i]\n i += 1\n chars: list[str] = []\n\n while i < n:\n ch = source[i]\n\n if ch == \"\\\\\":\n if i + 1 >= n:\n break\n chars.append(ch)\n chars.append(source[i + 1])\n i",
"cold_text": "```python\nimport re\n\ndef extract_regular_expressions(source: str) -> list[str]:\n \"\"\"\n Return every regular expression string literal used in the given Python source.\n\n This is a lexical scanner, not a full AST parser, so it handles normal string\n literals but does not evaluate them.\n \"\"\"\n expressions: list[str] = []\n i = 0\n n = len(source)\n\n while i < n:\n if source[i] not in \"'\\\"\":\n i += 1\n continue\n\n quote = source[i]\n i += 1\n chars: list[str] = []\n\n while i < n:\n ch = source[i]\n\n if ch == \"\\\\\":\n if i + 1 >= n:\n break\n chars.append(ch)\n chars.append(source[i + 1])\n i",
"warm_prompt_ms": 1003.44,
"cold_prompt_ms": 6170.8
},
{
"variant": 4,
"L": 7039,
"warm_prompt_n": 1028,
"warm_cache_n": 6011,
"cold_prompt_n": 7039,
"cold_cache_n": 0,
"aligned": true,
"warm_draft": [
null,
null
],
"cold_draft": [
null,
null
],
"identical": true,
"first_diff_char": null,
"warm_sha": "af97b2dde4e2",
"cold_sha": "af97b2dde4e2",
"instr": "Write a function that maps each class in this file to its base classes. ok",
"warm_instr": "Summarise this file. (warm-up 4) ok ok ok ok",
"warm_text": "```python\ndef map_classes_to_base_classes(source: str) -> dict[str, list[str]]:\n \"\"\"\n Return a mapping of class names to their direct base class names.\n\n Example:\n class A(B): pass\n class C(A): pass\n returns:\n {\"A\": [\"B\"], \"C\": [\"A\"]}\n \"\"\"\n import ast\n\n tree = ast.parse(source)\n result: dict[str, list[str]] = {}\n\n for node in tree.body:\n if isinstance(node, ast.ClassDef):\n bases = [ast.unparse(base) for base in node.bases]\n result[node.name] = bases\n\n return result\n```\n\nUsage:\n\n```python\nsource = open(\"file.py\", \"r\", encoding=\"utf-8\").read()\nprint(map_classes_to_base_classes(source))\n```\n\nFor the provided file",
"cold_text": "```python\ndef map_classes_to_base_classes(source: str) -> dict[str, list[str]]:\n \"\"\"\n Return a mapping of class names to their direct base class names.\n\n Example:\n class A(B): pass\n class C(A): pass\n returns:\n {\"A\": [\"B\"], \"C\": [\"A\"]}\n \"\"\"\n import ast\n\n tree = ast.parse(source)\n result: dict[str, list[str]] = {}\n\n for node in tree.body:\n if isinstance(node, ast.ClassDef):\n bases = [ast.unparse(base) for base in node.bases]\n result[node.name] = bases\n\n return result\n```\n\nUsage:\n\n```python\nsource = open(\"file.py\", \"r\", encoding=\"utf-8\").read()\nprint(map_classes_to_base_classes(source))\n```\n\nFor the provided file",
"warm_prompt_ms": 1005.61,
"cold_prompt_ms": 6177.63
}
],
"result": "PASS"
},
{
"label": "n-tools-q106",
"passed": 6,
"total": 14,
"detail": {
"multi-arg|think=True": false,
"nested-object|think=True": false,
"enum|think=True": false,
"correct-decline|think=True": false,
"multi-turn|think=True": false,
"streaming|think=True": false,
"parallel|think=True": false,
"multi-arg|think=False": true,
"nested-object|think=False": false,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-vision-q106-faon",
"fa": "on",
"mtp": false,
"expected": "red,blue,circle,square",
"answer": "The image shows two shapes: a red circle on the left and a blue square on the right.",
"hits": [
"red",
"blue",
"circle",
"square"
],
"error": null,
"server_died": false,
"server_log_errors": [],
"result": "PASS"
},
{
"label": "n-vision-q106-faoff",
"fa": "off",
"mtp": false,
"expected": "red,blue,circle,square",
"answer": "The image shows two simple shapes:\n\n- A **red circle** on the left.\n- A **blue square** on the right.",
"hits": [
"red",
"blue",
"circle",
"square"
],
"error": null,
"server_died": false,
"server_log_errors": [],
"result": "PASS"
},
{
"label": "n-tools-q106-tpl",
"passed": 14,
"total": 14,
"detail": {
"multi-arg|think=True": true,
"nested-object|think=True": true,
"enum|think=True": true,
"correct-decline|think=True": true,
"multi-turn|think=True": true,
"streaming|think=True": true,
"parallel|think=True": true,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-tools-q106-tpl-medium",
"passed": 7,
"total": 14,
"detail": {
"multi-arg|think=True": false,
"nested-object|think=True": false,
"enum|think=True": false,
"correct-decline|think=True": false,
"multi-turn|think=True": false,
"streaming|think=True": false,
"parallel|think=True": false,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-tools-q106-c1",
"passed": 13,
"total": 14,
"detail": {
"multi-arg|think=True": true,
"nested-object|think=True": false,
"enum|think=True": true,
"correct-decline|think=True": true,
"multi-turn|think=True": true,
"streaming|think=True": true,
"parallel|think=True": true,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-vision-q106-c1-faon",
"fa": "on",
"mtp": false,
"expected": "red,blue,circle,square",
"answer": "The image shows two shapes: a red circle on the left and a blue square on the right.",
"hits": [
"red",
"blue",
"circle",
"square"
],
"error": null,
"server_died": false,
"server_log_errors": [],
"result": "PASS"
},
{
"label": "n-tools-q106-roff",
"passed": 13,
"total": 14,
"detail": {
"multi-arg|think=True": true,
"nested-object|think=True": false,
"enum|think=True": true,
"correct-decline|think=True": true,
"multi-turn|think=True": true,
"streaming|think=True": true,
"parallel|think=True": true,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-tools-q106-roff-r2",
"passed": 13,
"total": 14,
"detail": {
"multi-arg|think=True": true,
"nested-object|think=True": true,
"enum|think=True": true,
"correct-decline|think=True": true,
"multi-turn|think=True": true,
"streaming|think=True": true,
"parallel|think=True": false,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-tools-q106-roff-r3",
"passed": 14,
"total": 14,
"detail": {
"multi-arg|think=True": true,
"nested-object|think=True": true,
"enum|think=True": true,
"correct-decline|think=True": true,
"multi-turn|think=True": true,
"streaming|think=True": true,
"parallel|think=True": true,
"multi-arg|think=False": true,
"nested-object|think=False": true,
"enum|think=False": true,
"correct-decline|think=False": true,
"multi-turn|think=False": true,
"streaming|think=False": true,
"parallel|think=False": true
}
},
{
"label": "n-vision-q106-roff-faon",
"fa": "on",
"mtp": false,
"expected": "red,blue,circle,square",
"answer": "The image shows two shapes: a red circle on the left and a blue square on the right.",
"hits": [
"red",
"blue",
"circle",
"square"
],
"error": null,
"server_died": false,
"server_log_errors": [],
"result": "PASS"
}
],
"sizing": [
{
"label": "strix-lean",
"ctx": 65536,
"avail_before": 122.34,
"footprint_loaded_gib": 21.11,
"footprint_after_8k_gib": 21.29
},
{
"label": "strix-lean",
"ctx": 262144,
"avail_before": 122.15,
"footprint_loaded_gib": 24.36,
"footprint_after_8k_gib": 24.52
}
],
"n_ubatch": 1024,
"template_fix": {
"file": "chat_template_enable_thinking.jinja",
"size_bytes": 7895,
"sha256": "9183c7ba8510fb9628edd2265a4cad8c02b3aecec3cbdf65620f47494e75836d",
"source_sha256": "f1753536417ee87cded4bc5017354eb2123c079d20c634403a3d808b2ba3cc5b",
"shim": "{%- if reasoning_effort is not defined and enable_thinking is defined %}{%- set reasoning_effort = 'high' if enable_thinking else 'none' %}{%- endif %}\n",
"server_flags": [
"--chat-template-file",
"chat_template_enable_thinking.jinja",
"--reasoning",
"off"
],
"probes_roff": {
"no-kwargs|correct-decline": {
"content": "391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"no-kwargs|single-word": {
"content": "ready",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"no-kwargs|multi-arg": {
"content": "",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": []
},
"enable_thinking=false|correct-decline": {
"content": "391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"enable_thinking=false|single-word": {
"content": "ready",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"enable_thinking=false|multi-arg": {
"content": "",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": []
},
"reasoning_effort=high|correct-decline": {
"content": "We need answer directly. 391.\n</think>\n\n391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": [
"</think>"
]
},
"reasoning_effort=high|single-word": {
"content": "We need need output exactly ready.\n</think>\n\nready",
"reasoning_len": 0,
"tool_calls": [],
"leaks": [
"</think>"
]
},
"reasoning_effort=high|multi-arg": {
"content": "We need need tool. Current weather Paris celsius.\n</think>\n\n",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": [
"</think>"
]
},
"reasoning_effort=medium|correct-decline": {
"content": "\n\n</think>\n\n391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": [
"</think>"
]
},
"reasoning_effort=medium|single-word": {
"content": "\n\n</think>\n\nready",
"reasoning_len": 0,
"tool_calls": [],
"leaks": [
"</think>"
]
},
"reasoning_effort=medium|multi-arg": {
"content": "\n\n</think>\n\n",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": [
"</think>"
]
},
"reasoning_effort=none|correct-decline": {
"content": "391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"reasoning_effort=none|single-word": {
"content": "ready",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"reasoning_effort=none|multi-arg": {
"content": "",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": []
}
},
"probes_high_default_on": {
"no-kwargs|correct-decline": {
"content": "391",
"reasoning_len": 30,
"tool_calls": [],
"leaks": []
},
"no-kwargs|multi-arg": {
"content": "",
"reasoning_len": 50,
"tool_calls": [
"get_weather"
],
"leaks": []
},
"reasoning_effort=medium|correct-decline": {
"content": "391",
"reasoning_len": 0,
"tool_calls": [],
"leaks": []
},
"reasoning_effort=medium|multi-arg": {
"content": "",
"reasoning_len": 0,
"tool_calls": [
"get_weather"
],
"leaks": []
},
"reasoning_effort=none|correct-decline": {
"content": "",
"reasoning_len": 3,
"tool_calls": [],
"leaks": []
},
"reasoning_effort=none|multi-arg": {
"content": "",
"reasoning_len": 133,
"tool_calls": [],
"leaks": []
}
},
"medium_mapping_label": "n-tools-q106-tpl-medium"
},
"tools_diag": {
"stock_on_replies": 7,
"stock_on_leaks": 7,
"stock_on_reasoning_extracted": 0,
"nested_off_attempts": 4,
"nested_off_http500": 2,
"gate_http500_logged": true,
"flag_probes": {
"default": {
"leaks": 3,
"n": 3
},
"fmt-deepseek": {
"leaks": 3,
"n": 3
},
"srv-kwargs-high": {
"leaks": 3,
"n": 3
},
"reasoning-on": {
"leaks": 3,
"n": 3
},
"tpl-enable-thinking": {
"leaks": 0,
"n": 3
}
}
},
"seats": {
"max1-nex-fast": {
"unit": "max1-nex-fast",
"port": 8097,
"load_s": 25,
"time": "2026-09-17T01:21:10Z",
"direct_reply": "ready",
"direct_tg": 41.41386950489719,
"default_reply": "ready",
"default_reasoning_len": 0,
"default_leak": false,
"thinking_reply": "",
"thinking_reasoning_len": 5,
"thinking_leak": false,
"gateway_model": "nex-n2.5-mini-fast@max1",
"gateway_reply": "ready",
"result": "PASS"
},
"max1-nex-fast-imat": {
"unit": "max1-nex-fast-imat",
"port": 8098,
"load_s": 25,
"time": "2026-09-17T01:21:43Z",
"direct_reply": "ready",
"direct_tg": 41.54290343352097,
"default_reply": "ready",
"default_reasoning_len": 0,
"default_leak": false,
"thinking_reply": "",
"thinking_reasoning_len": 5,
"thinking_leak": false,
"gateway_model": "nex-n2.5-mini-fast-imatrix@max1",
"gateway_reply": "ready",
"result": "PASS"
}
},
"measured": "2026-09-16",
"measured_range": [
"2026-09-16",
"2026-09-17"
]
}