{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18310",
    "created_at_utc": "2026-06-27T16:16:52.776758+00:00",
    "max_tokens": 512,
    "model": "gemma4-26b-a4b-q8",
    "prompt_chars": 3460,
    "prompt_mode": "filled-long",
    "prompt_preview": "You are running a deterministic Gemma B70 decode benchmark. Read the reference context, then produce a long numbered response until the token limit is reached. Do not summarize early.\n\nReference context:\nbenchmark latency memory throughput ",
    "prompt_sha256": "04f687a68d60254567d3252624e01a696b0c9843dbb02f5a789d75aada5e4da5",
    "prompt_sha256s": [
      "04f687a68d60254567d3252624e01a696b0c9843dbb02f5a789d75aada5e4da5",
      "04f687a68d60254567d3252624e01a696b0c9843dbb02f5a789d75aada5e4da5",
      "04f687a68d60254567d3252624e01a696b0c9843dbb02f5a789d75aada5e4da5"
    ],
    "prompt_tokens_requested": 512,
    "prompts_are_unique": false,
    "repeats": 3,
    "seed": 1,
    "usage_required": true
  },
  "bench_summary": {
    "completion_tokens": {
      "cv": 0.0,
      "max": 512,
      "mean": 512.0,
      "median": 512,
      "min": 512,
      "stdev": 0.0
    },
    "completion_tokens_total": 1536,
    "elapsed_s": {
      "cv": 0.05305984991699286,
      "max": 3.6750731610227376,
      "mean": 3.463588140012386,
      "median": 3.3730022039962932,
      "min": 3.342689055018127,
      "stdev": 0.18377746688333366
    },
    "prompt_tokens": {
      "cv": 0.0,
      "max": 588,
      "mean": 588.0,
      "median": 588,
      "min": 588,
      "stdev": 0.0
    },
    "requests": 3,
    "tok_s_after_ttft": {
      "cv": 0.005160937830055293,
      "max": 177.3917541664833,
      "mean": 176.40259133127742,
      "median": 176.21623213048554,
      "min": 175.59978769686344,
      "stdev": 0.9104028069213735
    },
    "tok_s_wall": {
      "cv": 0.05153392223686323,
      "max": 153.17009496631852,
      "mean": 148.0935352276448,
      "median": 151.7935563141312,
      "min": 139.3169544024847,
      "stdev": 7.6318407282036125
    },
    "ttft_s": {
      "cv": 0.3217662294490064,
      "max": 0.7695519510307349,
      "mean": 0.5610849846852943,
      "median": 0.45728113898076117,
      "min": 0.4564218640443869,
      "stdev": 0.18053819992264067
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 6144,
  "fresh_response_validity": {
    "all_cached_tokens": [
      0,
      0,
      0
    ],
    "all_prompt_hashes_distinct": false,
    "all_rows_fresh_response_mean_eligible": false,
    "cached_tokens_all_zero": true,
    "cached_tokens_reported": true,
    "headline_cached_tokens": 0,
    "headline_policy": "Use row0 only as fresh-response headline. Later repeated-prompt rows are support-only unless each request uses an independently fresh prompt with no reusable continuation/history.",
    "headline_row": 0,
    "headline_tok_s_after_ttft": 176.21623213048554,
    "headline_tok_s_wall": 139.3169544024847,
    "prompts_are_unique": false,
    "repeated_prompt_rows_support_only": true
  },
  "headline_eligible_for_gemma_q8": true,
  "label": "gemma4-q8-gpu0-q8reorder-vdr2-ub720-nmin3-pmin010-fullconfirm-20260627T155347Z",
  "launcher_identity": {
    "batch_size": "1024",
    "cache_type_k": null,
    "cache_type_v": null,
    "ctx_size": "8192",
    "extra_llama_args": "--parallel 1 --cache-ram 0 --spec-type draft-mtp --spec-draft-model /mnt/fast-ai/llm-models/gemma4-26b-a4b-it-q8-gguf/MTP/gemma-4-26B-A4B-it-Q4_0-MTP.gguf --spec-draft-n-max 7 --spec-draft-device SYCL0 --spec-draft-ngl all --spec-draft-type-k f16 --spec-draft-type-v f16 --spec-draft-n-min 3 --spec-draft-p-min 0.10 --no-spec-draft-backend-sampling --spec-draft-threads 32 --spec-draft-threads-batch 32 --ctx-checkpoints 0",
    "flash_attn": "off",
    "ggml_sycl_disable_dnn": "<unset>",
    "ggml_sycl_disable_graph": "0",
    "ggml_sycl_disable_opt": "0",
    "ggml_sycl_enable_vmm": "0",
    "gpu_index": "0",
    "llama_cpp_commit": "c926ad098",
    "llama_devices": "SYCL0",
    "llama_gemma4_fused_final_post_norm_residual": "<unset>",
    "llama_gemma4_moe_fused_branch_post_norm_add": "<unset>",
    "llama_gemma4_moe_fused_down_weighted_sum": "<unset>",
    "llama_gemma4_moe_fused_down_weighted_sum_debug": "<unset>",
    "llama_gemma4_moe_fused_down_weighted_sum_direct_f32_parallel_slots": "<unset>",
    "llama_gemma4_moe_fused_down_weighted_sum_matmul_epilogue": "<unset>",
    "llama_gemma4_moe_fused_down_weighted_sum_parallel_slots": "<unset>",
    "llama_gemma4_moe_fused_geglu_down_weighted_sum": "<unset>",
    "llama_gemma4_moe_fused_router_selected_weights": "<unset>",
    "llama_gemma4_moe_gateup_geglu": "<unset>",
    "llama_gemma4_moe_geglu_down_matmul_epilogue": "<unset>",
    "llama_gemma4_moe_reuse_attn_rms": "1",
    "llama_gemma4_moe_router_post_scale": "<unset>",
    "llama_gemma4_moe_selected_softmax": "1",
    "llama_gemma4_moe_selected_softmax_fused": "1",
    "llama_gemma4_moe_selected_softmax_weighted_sum": "<unset>",
    "llama_gemma4_moe_skip_early_weights_expand": "<unset>",
    "llama_gemma4_moe_sorted_top_k": "<unset>",
    "llama_gemma4_moe_top_k": "<unset>",
    "llama_gemma4_moe_weighted_sum": "1",
    "llama_gemma4_moe_weighted_sum_2d": "<unset>",
    "llama_gemma4_moe_weighted_sum_selected_softmax": "<unset>",
    "llama_gemma4_mtp_fused_output_argmax": "1",
    "llama_gemma4_mtp_qonly_attn_inputs": "1",
    "llama_main_gpu": "<unset>",
    "llama_mtp_defer_target_h_nextn": "1",
    "llama_mtp_draft_backend_argmax": "<unset>",
    "llama_mtp_draft_backend_topk": "<unset>",
    "llama_mtp_draft_device_h_handoff": "<unset>",
    "llama_mtp_draft_direct_argmax_ids": "1",
    "llama_mtp_draft_direct_argmax_scores": "<unset>",
    "llama_mtp_draft_direct_argmax_unroll": "7",
    "llama_mtp_draft_fast_argmax": "1",
    "llama_mtp_draft_fast_topk": "<unset>",
    "llama_mtp_draft_logit_gap_min": "<unset>",
    "llama_mtp_draft_profile": "<unset>",
    "llama_mtp_draft_terminal_logits_only": "<unset>",
    "llama_mtp_draft_top_k": "<unset>",
    "llama_server": "/home/steve/src/llama.cpp-gemma-record-repro-c926/build-sycl-b70-aot-bmg-g31-q8reorder-vdr2/bin/llama-server",
    "llama_server_spec_profile": "<unset>",
    "llama_spec_verify_backend_argmax_ids": "1",
    "llama_spec_verify_fused_output_argmax": "<unset>",
    "llama_spec_verify_greedy_argmax": "<unset>",
    "llama_spec_verify_softcap_argmax": null,
    "llama_split_mode": "<unset>",
    "llama_sycl_f16_p021_small_ncols": "<unset>",
    "llama_sycl_mul_mat_id_gate_up_fast_seed_route_cache": "<unset>",
    "llama_sycl_mul_mat_id_gate_up_q8_singleton_direct": "<unset>",
    "llama_sycl_mul_mat_id_multi_token_fast": "1",
    "llama_sycl_mul_mat_id_multi_token_filter": "<unset>",
    "llama_sycl_mul_mat_id_multi_token_grouped_q8_0": "<unset>",
    "llama_sycl_mul_mat_id_multi_token_name_substr": "<unset>",
    "llama_sycl_mul_mat_id_multi_token_no_reorder": "<unset>",
    "llama_sycl_mul_mat_id_multi_token_per_slot_q8_0": "<unset>",
    "llama_sycl_mul_mat_id_q8_0_reorder": "1",
    "llama_sycl_mul_mat_id_route_cache": "1",
    "llama_sycl_mul_mat_id_route_cache_device_map": "<unset>",
    "llama_sycl_mul_mat_id_route_cache_inplace": "<unset>",
    "llama_sycl_mul_mat_id_route_timing": "<unset>",
    "llama_sycl_mul_mat_id_route_timing_every": "<unset>",
    "llama_sycl_q8_mmvq_small_ncols": "<unset>",
    "llama_tensor_split": "<unset>",
    "oneapi_device_selector": "level_zero:0",
    "poll": "100",
    "port": "18310",
    "reasoning": "off",
    "threads": "8",
    "ubatch_size": "720"
  },
  "model_file_bytes": 27636230944,
  "model_path": "/mnt/fast-ai/llm-models/gemma4-26b-a4b-it-q8-gguf/gemma-4-26B-A4B-it-UD-Q8_K_XL.gguf",
  "quality_lane": "q8_target",
  "run_dir": "/home/steve/qwen36-results-main/data/gemma4-q8-gpu0-q8reorder-vdr2-ub720-nmin3-pmin010-fullconfirm-20260627T155347Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu0-q8reorder-vdr2-ub720-nmin3-pmin010-fullconfirm-20260627T155347Z.server.log"
}
