{ "model_id": "m5", "model_tag": "gemma4:26b", "box_class": "gpu-3080", "arm": "cpu-box", "arm_label": "cpu-box (CPU-only on the same box, size_vram == 0)", "base_url": "http://127.0.0.1:11439", "started_utc": "2026-09-10T12:48:16Z", "nproc": 20, "num_predict": 256, "prompt_path": "withheld-2026-11-10", "prompt_sha256": "90eedd0c53f9554ae3837674504fcb7090013d9a432a653352183c0f25a7ce5c", "prompt_bytes": 2101, "scored_runs_requested": 3, "warmup_discarded_runs": 1, "num_gpu_option": null, "wall_decode_note": "decode_tok_s is ollama's own eval_count/eval_duration. decode_tok_s_wall is (eval_count - 1) / (wall_s - ttft_s): the tokens AFTER the first, over the time after the first, measured by this client's clock -- it carries the HTTP hop and the stream decode that ollama's counter does not. decode_tok_s_wall_gross is eval_count / wall_s and therefore includes prefill and any load; it is the slowest and least comparable of the three, printed so nobody has to derive it.", "loadavg_before_arm": "10.10 9.74 8.01 2/807 981711", "free_m_before": "total used free shared buff/cache available\nMem: 47040 3720 15867 9 27986 43320\nSwap: 24575 969 23606", "runs": [ { "ttft_ms": 752.2, "first_token_was_thinking": false, "wall_s": 20.7524, "eval_count": 256, "eval_duration_ns": 19997438000, "decode_tok_s": 12.802, "prompt_eval_count": 536, "prompt_eval_duration_ns": 286401000, "prefill_tok_s": 1871.502, "load_duration_ns": 461596672, "total_duration_ns": 20749377555, "streamed_chunks": 256, "response_chars": 1045, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 12.75, "decode_tok_s_wall_gross": 12.336, "run": 1 }, { "ttft_ms": 747.85, "first_token_was_thinking": false, "wall_s": 20.7323, "eval_count": 256, "eval_duration_ns": 19981827000, "decode_tok_s": 12.812, "prompt_eval_count": 536, "prompt_eval_duration_ns": 285036000, "prefill_tok_s": 1880.464, "load_duration_ns": 458620874, "total_duration_ns": 20729367016, "streamed_chunks": 256, "response_chars": 1045, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 12.76, "decode_tok_s_wall_gross": 12.348, "run": 2 }, { "ttft_ms": 752.17, "first_token_was_thinking": false, "wall_s": 20.7556, "eval_count": 256, "eval_duration_ns": 20000807000, "decode_tok_s": 12.799, "prompt_eval_count": 536, "prompt_eval_duration_ns": 286111000, "prefill_tok_s": 1873.399, "load_duration_ns": 461879225, "total_duration_ns": 20752648303, "streamed_chunks": 256, "response_chars": 1045, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 12.748, "decode_tok_s_wall_gross": 12.334, "run": 3 } ], "errors": [], "nvidia_before": "0, , NVIDIA GeForce RTX 3080, 256, 11.33, 380.00, 25, 210", "smi_before": [ { "index": "0", "uuid": "", "name": "NVIDIA GeForce RTX 3080", "memory.used": 256.0, "power.draw": 11.33, "power.limit": 380.0, "temperature.gpu": 25.0, "clocks.sm": 210.0 } ], "ollama_version": "0.32.13", "quantization": "Q4_K_M", "param_size": "25.8B", "family": "gemma4", "context_length": 262144, "think_mode": false, "think_note": "think:false accepted", "warmup_discarded": { "ttft_ms": 746.3, "first_token_was_thinking": false, "wall_s": 20.6326, "eval_count": 256, "eval_duration_ns": 19883628000, "decode_tok_s": 12.875, "prompt_eval_count": 536, "prompt_eval_duration_ns": 280869000, "prefill_tok_s": 1908.363, "load_duration_ns": 461308054, "total_duration_ns": 20629661073, "streamed_chunks": 256, "response_chars": 1045, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 12.823, "decode_tok_s_wall_gross": 12.408 }, "ps_after_load": [ { "name": "gemma4:26b", "model": "gemma4:26b", "size": 17953655356, "digest": "5571076f3d70050487b26b341705799e0ab29b808164f90d20d4cf84f699d251", "details": { "parent_model": "", "format": "gguf", "family": "gemma4", "families": [ "gemma4" ], "parameter_size": "25.8B", "quantization_level": "Q4_K_M" }, "expires_at": "2026-09-10T12:58:59.447540526Z", "size_vram": 0, "context_length": 4096 } ], "size_total": 17953655356, "size_vram": 0, "gpu": { "arm": "cpu-box", "requested_uuid": "", "requested_index": null, "fits_on_card": false, "card_untouched": true, "size_bytes": 17953655356, "size_vram_bytes": 0, "offload_fraction": 0.0, "offload_split": "0.0% VRAM / 100.0% RAM", "verdict": "cpu-only: size_vram == 0, nothing on the card", "smi_before": [ { "index": "0", "uuid": "", "name": "NVIDIA GeForce RTX 3080", "memory.used": 256.0, "power.draw": 11.33, "power.limit": 380.0, "temperature.gpu": 25.0, "clocks.sm": 210.0 } ], "during": { "sample_hz": 1.0, "samples": 61, "errors": [], "per_gpu": { "": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 10.96, "median": 11.17, "max": 11.54, "mean": 11.18, "n": 61 }, "power_limit_w": 380.0, "memory_used_mib": { "min": 256.0, "median": 256.0, "max": 256.0, "mean": 256.0, "n": 61 }, "temperature_c": { "min": 25.0, "median": 25.0, "max": 25.0, "mean": 25.0, "n": 61 }, "clocks_sm_mhz": { "min": 210.0, "median": 210.0, "max": 210.0, "mean": 210.0, "n": 61 }, "n": 61 } }, "target_uuid": "", "target": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 10.96, "median": 11.17, "max": 11.54, "mean": 11.18, "n": 61 }, "power_limit_w": 380.0, "memory_used_mib": { "min": 256.0, "median": 256.0, "max": 256.0, "mean": 256.0, "n": 61 }, "temperature_c": { "min": 25.0, "median": 25.0, "max": 25.0, "mean": 25.0, "n": 61 }, "clocks_sm_mhz": { "min": 210.0, "median": 210.0, "max": 210.0, "mean": 210.0, "n": 61 }, "n": 61 } }, "smi_after": [ { "index": "0", "uuid": "", "name": "NVIDIA GeForce RTX 3080", "memory.used": 256.0, "power.draw": 11.12, "power.limit": 380.0, "temperature.gpu": 25.0, "clocks.sm": 210.0 } ] }, "scored": true, "peak_runner_rss_kb": 19007072, "peak_runner_rss_gib": 18.13, "peak_runner_threads": 32, "rss_samples": 121, "loadavg_during_runs": { "min": 8.96, "median": 9.43, "max": 9.74, "mean": 9.44, "samples": 121, "nproc": 20 }, "contention_suspected": false, "nvidia_after": "0, , NVIDIA GeForce RTX 3080, 256, 11.12, 380.00, 25, 210", "smi_after": [ { "index": "0", "uuid": "", "name": "NVIDIA GeForce RTX 3080", "memory.used": 256.0, "power.draw": 11.12, "power.limit": 380.0, "temperature.gpu": 25.0, "clocks.sm": 210.0 } ], "free_m_after": "total used free shared buff/cache available\nMem: 47040 22319 422 9 24832 24721\nSwap: 24575 969 23606", "derived": { "decode_tok_s": { "min": 12.799, "median": 12.802, "max": 12.812, "mean": 12.804, "n": 3 }, "decode_tok_s_wall": { "min": 12.748, "median": 12.75, "max": 12.76, "mean": 12.753, "n": 3 }, "ttft_ms": { "min": 747.85, "median": 752.17, "max": 752.2, "mean": 750.74, "n": 3 }, "wall_s": { "min": 20.7323, "median": 20.7524, "max": 20.7556, "mean": 20.7468, "n": 3 }, "prefill_tok_s": { "min": 1871.502, "median": 1873.399, "max": 1880.464, "mean": 1875.122, "n": 3 }, "eval_count": { "min": 256, "median": 256, "max": 256, "mean": 256.0, "n": 3 }, "thinking_chars_total": 0 }, "ttft_note": "Every scored run sends the SAME frozen prompt, so from the second request the KV prefix is cached and derived.ttft_ms is a CACHED-PREFIX first-token time, not a fresh-prompt one. LEDGER.md, 'THE HEADLINE TRAP'.", "ps_final": [], "finished_utc": "2026-09-10T12:50:02Z" }