{"created_utc": "2026-10-09T13:01:12.124260+00:00", "engines": [{"binary": "/tmp/claude-1000/-home-germar-workplace-geistlib/108234fc-8402-4734-b735-41f3bce6f71b/scratchpad/wt/p6-cpu/bin/linux/release/tests/bench_perf_sweep", "binary_sha256": "1725aab44f344ed21833bd48223a5bdc6ff0d5330e0029bff2d7e04cfc61de35", "commit": "8489cb9537749f11e775267fc6ab5ce65b4b845f", "kind": "geist", "label": "geist", "linked_libraries": ["linux-vdso.so.1 (0x000071bfb94a2000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x000071bfb91cc000)", "\tlibmvec.so.1 => /usr/lib/x86_64-linux-gnu/libmvec.so.1 (0x000071bfb90d1000)", "\tlibopenblas.so.0 => /usr/lib/x86_64-linux-gnu/libopenblas.so.0 (0x000071bfb6bf8000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x000071bfb6b99000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x000071bfb6800000)", "\t/lib64/ld-linux-x86-64.so.2 (0x000071bfb94a4000)", "\tlibgfortran.so.5 => /usr/lib/x86_64-linux-gnu/libgfortran.so.5 (0x000071bfb6400000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x000071bfb6b69000)"]}, {"binary": "/home/germar/llama.cpp-v0.6.0/build-cpu/bin/llama-bench", "binary_sha256": "b300abddc3321e66fcffdedb39c2d03b11c72e1e9abb48d9a4c3c1b866c83078", "commit": "d81235049384534c167caea52b85a694f6103d14", "kind": "llama", "label": "llama.cpp", "linked_libraries": ["linux-vdso.so.1 (0x000070194307b000)", "\tlibllama-bench-impl.so => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libllama-bench-impl.so (0x0000701943008000)", "\tlibstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x0000701942c00000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x0000701942fbd000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x0000701942800000)", "\tlibllama-common.so.0 => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libllama-common.so.0 (0x0000701942200000)", "\tlibllama.so.0 => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libllama.so.0 (0x0000701941c00000)", "\tlibggml.so.0 => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libggml.so.0 (0x0000701942fad000)", "\tlibggml-base.so.0 => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libggml-base.so.0 (0x0000701942ed2000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x0000701942ada000)", "\t/lib64/ld-linux-x86-64.so.2 (0x000070194307d000)", "\tlibssl.so.3 => /usr/lib/x86_64-linux-gnu/libssl.so.3 (0x00007019420ef000)", "\tlibcrypto.so.3 => /usr/lib/x86_64-linux-gnu/libcrypto.so.3 (0x0000701941400000)", "\tlibggml-cpu.so.0 => /home/germar/llama.cpp-v0.6.0/build-cpu/bin/libggml-cpu.so.0 (0x0000701941000000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x0000701942a7b000)", "\tlibz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1 (0x0000701942eb2000)", "\tlibzstd.so.1 => /usr/lib/x86_64-linux-gnu/libzstd.so.1 (0x000070194273d000)"]}], "host_profile": "amd_9950x", "kind": "metadata", "model": {"file": "bitnet_b1_58-large-TQ2_0.gguf", "key": "bitnet-large-tq2", "path": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "sha256": "281aafb18a9f4a3124c10a1d8683e2296f0cfe8a2944da0a5667d17488a951bb"}, "protocol": {"host_gate": {"cooldown_seconds": 60, "max_load_per_core": 0.2, "settle_seconds": 30, "temp_gate": {"max_millic": 62000, "note": "passively cooled hosts (pi5) throttle at ~80C; measurements start below 62C so a full phase stays under throttle. Hosts without the sensor path skip this gate.", "sensor": "/sys/class/thermal/thermal_zone0/temp"}, "timeout_seconds": 14400}, "host_profiles": {"amd_9950x": {"decode_threads": 15, "expect_backend": "cpu_x86", "note": "16 physical cores, SMT siblings excluded. Measured, not assumed: benchmark/results/THREAD-SWEEP-AMD9950X.md sweeps 4-32 threads and finds the joint optimum at 16 for both engines, with llama.cpp prefill collapsing ~6x under SMT. Decode uses 15 by the core-minus-one rule; that specific value sits on the 12-16 plateau but was not measured on its own.", "prefill_threads": 16}, "apple_m1max": {"decode_threads": 7, "expect_backend": "cpu_neon", "note": "8 performance cores; decode leaves one core to the OS", "prefill_threads": 8}, "pi5": {"decode_threads": 3, "expect_backend": "cpu_neon", "note": "4x Cortex-A76, 4 GB reference board; decode leaves one core to the OS (same rule as apple_m1max). Thermals: active-cooling assumption; runner host gate handles load/settle.", "prefill_threads": 4}}, "models": {"bitnet-large-tq2": {"file": "bitnet_b1_58-large-TQ2_0.gguf", "sha256": "281aafb18a9f4a3124c10a1d8683e2296f0cfe8a2944da0a5667d17488a951bb"}, "bonsai2-27b-pq2": {"file": "Ternary-Bonsai-2-27B-PQ2_0.gguf", "sha256": "3907dc1658db1f78a9826bf8d5bcb8dc65db0d466388937af57f2294fae62ec1"}, "gemma4-e2b-q4km": {"file": "gemma4-e2b-Q4_K_M.gguf", "sha256": "740185b21d22ceb83a11c3aa62ad5842ef32c70f6096d756bbee85a1e4ec34b8"}, "gemma4-e4b-q4km": {"file": "gemma4-e4b-Q4_K_M.gguf", "sha256": "85a896a047553e842f25297ee5b031d64ff30147d9c4af17b1e4b394cd1fab87"}, "llama32-3b-q4km": {"file": "llama-3.2-3b-Q4_K_M.gguf", "sha256": "6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff"}, "qwen3-0.6b-q8": {"file": "qwen3-0.6b-q8_0.gguf", "sha256": "9465e63a22add5354d9bb4b99e90117043c7124007664907259bd16d043bb031"}, "qwen35-0.8b-q8": {"file": "qwen3.5-0.8b-q8_0.gguf", "sha256": "0ad885ffd4bb022fc4f0d33a3308fa108ef8613159d3b3a67e23abca056b7a6c"}, "qwen35-4b-q4": {"file": "qwen3.5-4b-q4_0.gguf", "sha256": "298fcb5fe7a77ccc79745ae24751560c5ac56874caff4bb39b1f2055bd72b8bb"}, "qwen38-27b-q4": {"file": "qwen3.8-27b-q4_0.gguf", "sha256": "ede16c7b36e578ca87a8c70e011e4b4633a32c831c0ce76d0f474582384e671d"}}, "schema": "geist.benchmark.cross-engine-cpu.v2", "workload": {"affinity": "thread-count-only; operating-system scheduled for both engines", "aggregation": "median", "backend": "cpu-only", "cycles": 4, "decode_n": 64, "dispersion": "median_absolute_deviation", "kv_cache": "f16", "model_load_included": false, "repeats": 3, "seq_lens": [128, 256, 512, 1024], "thread_policy": "per host_profile; identical for both engines within a run", "thread_polling": "active", "warmup": "one discarded engine-native warmup before each measured test"}}, "protocol_sha256": "797d821c63d80cffb7e9496b561b10614b61d584210e5f67a0b8844b56e3325d", "runner": {"commit": "53aacaacb9d4b51c2c860720c016b8b5ea5322ec", "path": "tools/bench_cross_engine.py", "sha256": "1b2af101ff1c7bc98722fb96a3f43dff448df3808c9012cc7d69475f04c067e5"}, "schema": "geist.benchmark.cross-engine-run.v1", "system": {"compiler": "cc (Ubuntu 15.2.0-16ubuntu1) 15.2.0", "cpu": "AMD Ryzen 9 9950X 16-Core Processor", "logical_cores": 32, "memory_bytes": "63813259264", "os": "Linux 7.0.0-34-generic", "performance_cores": "16", "product": "Ubuntu 26.04.1 LTS"}}
{"cycle": 0, "elapsed_seconds": 5.638168195029721, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 2.50341796875, "load_per_core": 0.0782318115234375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.27 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          15.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.4\u00b0C  ", "Tccd1:        +35.6\u00b0C  ", "Tccd2:        +36.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 1.328125, "load_per_core": 0.04150390625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.36 V  ", "vddnb:         1.19 V  ", "edge:         +37.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.6\u00b0C  ", "Tccd1:        +26.8\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 0, "rows": [{"decode_tps": [319.8288915430245, 320.5320832582086, 318.52681348761973], "native": {"agg": "mean", "decode_ms": 200.23, "decode_ms_best": 199.67, "decode_ms_worst": 200.92, "decode_n": 64, "decode_tps": 319.627, "prefill_ms": 55.06, "prefill_ms_best": 53.29, "prefill_ms_worst": 57.97, "prefill_tps": 2324.904, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.51, "samples": {"decode_ms": [200.107, 199.668, 200.925], "prefill_ms": [53.907, 53.293, 57.968], "total_ms": [254.014, 252.961, 258.893], "ttft_ms": [53.908, 53.293, 57.968]}, "seq_len": 128, "threads": 16, "total_ms": 255.29, "total_ms_best": 252.96, "total_ms_worst": 258.89, "total_tps": 752.088, "total_tps_best": 759.01, "total_tps_worst": 741.62, "ttft_ms": 55.06, "warmup": 64}, "prefill_tps": [2374.4597176618995, 2401.8163736325596, 2208.1148219707425], "seq_len": 128}, {"decode_tps": [300.850843792601, 302.72929378931934, 301.52553073204746], "native": {"agg": "mean", "decode_ms": 212.13, "decode_ms_best": 211.41, "decode_ms_worst": 212.73, "decode_n": 64, "decode_tps": 301.7, "prefill_ms": 108.79, "prefill_ms_best": 107.47, "prefill_ms_worst": 109.56, "prefill_tps": 2353.162, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.51, "samples": {"decode_ms": [212.73, 211.41, 212.254], "prefill_ms": [109.56, 107.47, 109.339], "total_ms": [322.29, 318.88, 321.593], "ttft_ms": [109.561, 107.471, 109.339]}, "seq_len": 256, "threads": 16, "total_ms": 320.92, "total_ms_best": 318.88, "total_ms_worst": 322.29, "total_tps": 997.13, "total_tps_best": 1003.511, "total_tps_worst": 992.894, "ttft_ms": 108.79, "warmup": 64}, "prefill_tps": [2336.6192040890837, 2382.0601097980834, 2341.342064588116], "seq_len": 256}, {"decode_tps": [268.8454348784945, 267.9090950951705, 268.27969835301417], "native": {"agg": "mean", "decode_ms": 238.5, "decode_ms_best": 238.05, "decode_ms_worst": 238.89, "decode_n": 64, "decode_tps": 268.345, "prefill_ms": 234.29, "prefill_ms_best": 232.98, "prefill_ms_worst": 236.53, "prefill_tps": 2185.286, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.51, "samples": {"decode_ms": [238.055, 238.887, 238.557], "prefill_ms": [232.976, 236.528, 233.379], "total_ms": [471.031, 475.414, 471.936], "ttft_ms": [232.977, 236.528, 233.379]}, "seq_len": 512, "threads": 16, "total_ms": 472.79, "total_ms_best": 471.03, "total_ms_worst": 475.41, "total_tps": 1218.291, "total_tps_best": 1222.85, "total_tps_worst": 1211.575, "ttft_ms": 234.29, "warmup": 64}, "prefill_tps": [2197.6512602156445, 2164.648582831631, 2193.8563452581425], "seq_len": 512}, {"decode_tps": [221.0669936615948, 219.23138845404193, 220.552000303259], "native": {"agg": "mean", "decode_ms": 290.54, "decode_ms_best": 289.5, "decode_ms_worst": 291.93, "decode_n": 64, "decode_tps": 220.281, "prefill_ms": 513.71, "prefill_ms_best": 513.63, "prefill_ms_worst": 513.78, "prefill_tps": 1993.356, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.51, "samples": {"decode_ms": [289.505, 291.929, 290.181], "prefill_ms": [513.777, 513.709, 513.634], "total_ms": [803.282, 805.637, 803.815], "ttft_ms": [513.777, 513.709, 513.634]}, "seq_len": 1024, "threads": 16, "total_ms": 804.24, "total_ms_best": 803.28, "total_ms_worst": 805.64, "total_tps": 1352.822, "total_tps_best": 1354.444, "total_tps_worst": 1350.484, "ttft_ms": 513.71, "warmup": 64}, "prefill_tps": [1993.0826019849076, 1993.3464276467807, 1993.6374928450998], "seq_len": 1024}], "stderr": ["[bench] backend: cpu_x86", "[bench] warmup phase (discarded): 64-token prefill + 4 decode steps", "[bench] measuring 3 repeat(s) per seq_len; reporting the MEAN (with best/worst)"]}
{"cycle": 0, "elapsed_seconds": 11.65135010983795, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 2.841796875, "load_per_core": 0.08880615234375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.8\u00b0C  ", "Tccd1:        +41.5\u00b0C  ", "Tccd2:        +34.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.54541015625, "load_per_core": 0.0170440673828125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.07 V  ", "vddnb:         1.19 V  ", "edge:         +37.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.4\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [231.757, 231.922, 228.891], "native": {"decode": {"avg_ns": 277238389, "avg_ts": 230.856683, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [276151352, 275954983, 279608833], "samples_ts": [231.757, 231.922, 228.891], "split_mode": "layer", "stddev_ns": 2055278, "stddev_ts": 1.704158, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:44Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 93975952, "avg_ts": 1362.111292, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 128, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [94782234, 93894715, 93250907], "samples_ts": [1350.46, 1363.23, 1372.64], "split_mode": "layer", "stddev_ns": 768888, "stddev_ts": 11.130546, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1350.46, 1363.23, 1372.64], "seq_len": 128}, {"decode_tps": [213.532, 216.52, 215.192], "native": {"decode": {"avg_ns": 297571656, "avg_ts": 215.0812, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [299721024, 295584922, 297409022], "samples_ts": [213.532, 216.52, 215.192], "split_mode": "layer", "stddev_ns": 2072841, "stddev_ts": 1.49704, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:44Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 181943841, "avg_ts": 1407.037424, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 256, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [182211326, 182351912, 181268285], "samples_ts": [1404.96, 1403.88, 1412.27], "split_mode": "layer", "stddev_ns": 589256, "stddev_ts": 4.564927, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1404.96, 1403.88, 1412.27], "seq_len": 256}, {"decode_tps": [181.447, 184.695, 184.218], "native": {"decode": {"avg_ns": 348884070, "avg_ts": 183.453259, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [352720406, 346517473, 347414333], "samples_ts": [181.447, 184.695, 184.218], "split_mode": "layer", "stddev_ns": 3352594, "stddev_ts": 1.753895, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:46Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 374217407, "avg_ts": 1368.198509, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 512, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [374601947, 375210174, 372840100], "samples_ts": [1366.78, 1364.57, 1373.24], "split_mode": "layer", "stddev_ns": 1230941, "stddev_ts": 4.506797, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:39Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1366.78, 1364.57, 1373.24], "seq_len": 512}, {"decode_tps": [140.803, 145.772, 146.668], "native": {"decode": {"avg_ns": 443312053, "avg_ts": 144.414445, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [454534906, 439041036, 436360219], "samples_ts": [140.803, 145.772, 146.668], "split_mode": "layer", "stddev_ns": 9811314, "stddev_ts": 3.159265, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:47Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 772328392, "avg_ts": 1325.862732, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 1024, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [772969475, 771037019, 772978684], "samples_ts": [1324.76, 1328.08, 1324.75], "split_mode": "layer", "stddev_ns": 1119062, "stddev_ts": 1.921525, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:04:40Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1324.76, 1328.08, 1324.75], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 1, "elapsed_seconds": 11.671921147033572, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 2.99853515625, "load_per_core": 0.0937042236328125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.36 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          18.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.1\u00b0C  ", "Tccd1:        +37.5\u00b0C  ", "Tccd2:        +35.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.72314453125, "load_per_core": 0.0225982666015625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.2\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +26.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [226.654, 228.869, 229.429], "native": {"decode": {"avg_ns": 280319521, "avg_ts": 228.317195, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [282368175, 279636415, 278953975], "samples_ts": [226.654, 228.869, 229.429], "split_mode": "layer", "stddev_ns": 1806855, "stddev_ts": 1.466951, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:20Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 94295278, "avg_ts": 1357.459594, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 128, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [94823115, 94087462, 93975258], "samples_ts": [1349.88, 1360.44, 1362.06], "split_mode": "layer", "stddev_ns": 460652, "stddev_ts": 6.612601, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:25Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1349.88, 1360.44, 1362.06], "seq_len": 128}, {"decode_tps": [212.498, 220.3, 219.658], "native": {"decode": {"avg_ns": 294351568, "avg_ts": 217.485231, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [301179807, 290513330, 291361567], "samples_ts": [212.498, 220.3, 219.658], "split_mode": "layer", "stddev_ns": 5928618, "stddev_ts": 4.331263, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:20Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 184929427, "avg_ts": 1384.414139, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 256, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [184664422, 183126765, 186997094], "samples_ts": [1386.3, 1397.94, 1369.01], "split_mode": "layer", "stddev_ns": 1948725, "stddev_ts": 14.558502, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1386.3, 1397.94, 1369.01], "seq_len": 256}, {"decode_tps": [184.396, 186.193, 186.839], "native": {"decode": {"avg_ns": 344449768, "avg_ts": 185.809346, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [347079093, 343729447, 342540764], "samples_ts": [184.396, 186.193, 186.839], "split_mode": "layer", "stddev_ns": 2353349, "stddev_ts": 1.265902, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:22Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 374277557, "avg_ts": 1367.978539, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 512, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [375118717, 374842395, 372871559], "samples_ts": [1364.9, 1365.91, 1373.13], "split_mode": "layer", "stddev_ns": 1225443, "stddev_ts": 4.486978, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1364.9, 1365.91, 1373.13], "seq_len": 512}, {"decode_tps": [142.227, 144.833, 145.746], "native": {"decode": {"avg_ns": 443665048, "avg_ts": 144.268458, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [449985171, 441888817, 439121156], "samples_ts": [142.227, 144.833, 145.746], "split_mode": "layer", "stddev_ns": 5645613, "stddev_ts": 1.825996, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:23Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 777018236, "avg_ts": 1317.927944, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 1024, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [784931009, 772032805, 774090894], "samples_ts": [1304.57, 1326.37, 1322.84], "split_mode": "layer", "stddev_ns": 6929496, "stddev_ts": 11.699135, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:06:28Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1304.57, 1326.37, 1322.84], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 1, "elapsed_seconds": 5.798058690037578, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.95556640625, "load_per_core": 0.0611114501953125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.23 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          16.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.2\u00b0C  ", "Tccd1:        +35.6\u00b0C  ", "Tccd2:        +41.0\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.73291015625, "load_per_core": 0.0229034423828125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.07 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:           8.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.2\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +28.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 1, "rows": [{"decode_tps": [294.4302085393961, 302.931333986529, 301.22466653488084], "native": {"agg": "mean", "decode_ms": 213.7, "decode_ms_best": 211.27, "decode_ms_worst": 217.37, "decode_n": 64, "decode_tps": 299.483, "prefill_ms": 67.53, "prefill_ms_best": 62.6, "prefill_ms_worst": 74.82, "prefill_tps": 1895.47, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.54, "samples": {"decode_ms": [217.369, 211.269, 212.466], "prefill_ms": [65.162, 62.602, 74.824], "total_ms": [282.531, 273.871, 287.29], "ttft_ms": [65.163, 62.603, 74.824]}, "seq_len": 128, "threads": 16, "total_ms": 281.23, "total_ms_best": 273.87, "total_ms_worst": 287.29, "total_tps": 682.713, "total_tps_best": 701.06, "total_tps_worst": 668.315, "ttft_ms": 67.53, "warmup": 64}, "prefill_tps": [1964.3350418955831, 2044.6631098047988, 1710.6810648989629], "seq_len": 128}, {"decode_tps": [289.7094304946336, 290.11259995285667, 303.5577921862327], "native": {"agg": "mean", "decode_ms": 217.45, "decode_ms_best": 210.83, "decode_ms_worst": 220.91, "decode_n": 64, "decode_tps": 294.321, "prefill_ms": 138.52, "prefill_ms_best": 110.66, "prefill_ms_worst": 156.08, "prefill_tps": 1848.102, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.54, "samples": {"decode_ms": [220.911, 220.604, 210.833], "prefill_ms": [148.827, 156.077, 110.657], "total_ms": [369.738, 376.681, 321.49], "ttft_ms": [148.828, 156.077, 110.658]}, "seq_len": 256, "threads": 16, "total_ms": 355.97, "total_ms_best": 321.49, "total_ms_worst": 376.68, "total_tps": 898.952, "total_tps_best": 995.364, "total_tps_worst": 849.525, "ttft_ms": 138.52, "warmup": 64}, "prefill_tps": [1720.1179893433316, 1640.2160472074681, 2313.45509095674], "seq_len": 256}, {"decode_tps": [270.21777864097345, 270.1436404388146, 270.5776409857482], "native": {"agg": "mean", "decode_ms": 236.76, "decode_ms_best": 236.53, "decode_ms_worst": 236.91, "decode_n": 64, "decode_tps": 270.313, "prefill_ms": 230.99, "prefill_ms_best": 229.27, "prefill_ms_worst": 232.58, "prefill_tps": 2216.57, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.54, "samples": {"decode_ms": [236.846, 236.911, 236.531], "prefill_ms": [229.267, 231.111, 232.584], "total_ms": [466.113, 468.022, 469.115], "ttft_ms": [229.268, 231.112, 232.585]}, "seq_len": 512, "threads": 16, "total_ms": 467.75, "total_ms_best": 466.11, "total_ms_worst": 469.12, "total_tps": 1231.426, "total_tps_best": 1235.751, "total_tps_worst": 1227.843, "ttft_ms": 230.99, "warmup": 64}, "prefill_tps": [2233.2040808315196, 2215.385680473885, 2201.3552093007256], "seq_len": 512}, {"decode_tps": [220.01519479939083, 220.21353831543524, 219.2411515641486], "native": {"agg": "mean", "decode_ms": 291.14, "decode_ms_best": 290.63, "decode_ms_worst": 291.92, "decode_n": 64, "decode_tps": 219.822, "prefill_ms": 511.53, "prefill_ms_best": 510.38, "prefill_ms_worst": 512.45, "prefill_tps": 2001.83, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.54, "samples": {"decode_ms": [290.889, 290.627, 291.916], "prefill_ms": [510.38, 512.45, 511.765], "total_ms": [801.27, 803.077, 803.682], "ttft_ms": [510.381, 512.451, 511.766]}, "seq_len": 1024, "threads": 16, "total_ms": 802.68, "total_ms_best": 801.27, "total_ms_worst": 803.68, "total_tps": 1355.465, "total_tps_best": 1357.845, "total_tps_worst": 1353.77, "ttft_ms": 511.53, "warmup": 64}, "prefill_tps": [2006.3482111368, 1998.2437310957164, 2000.9183902767872], "seq_len": 1024}], "stderr": ["[bench] backend: cpu_x86", "[bench] warmup phase (discarded): 64-token prefill + 4 decode steps", "[bench] measuring 3 repeat(s) per seq_len; reporting the MEAN (with best/worst)"]}
{"cycle": 2, "elapsed_seconds": 5.648541735950857, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 2.77001953125, "load_per_core": 0.0865631103515625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.27 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          15.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.0\u00b0C  ", "Tccd1:        +35.6\u00b0C  ", "Tccd2:        +35.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.453125, "load_per_core": 0.01416015625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.06 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.4\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +28.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 0, "rows": [{"decode_tps": [311.24165969615035, 320.88403551785166, 321.27385081849536], "native": {"agg": "mean", "decode_ms": 201.43, "decode_ms_best": 199.21, "decode_ms_worst": 205.63, "decode_n": 64, "decode_tps": 317.732, "prefill_ms": 59.25, "prefill_ms_best": 53.85, "prefill_ms_worst": 69.65, "prefill_tps": 2160.447, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.56, "samples": {"decode_ms": [205.628, 199.449, 199.207], "prefill_ms": [69.651, 53.853, 54.237], "total_ms": [275.278, 253.302, 253.444], "ttft_ms": [69.651, 53.854, 54.238]}, "seq_len": 128, "threads": 16, "total_ms": 260.67, "total_ms_best": 253.3, "total_ms_worst": 275.28, "total_tps": 736.55, "total_tps_best": 757.987, "total_tps_worst": 697.476, "ttft_ms": 59.25, "warmup": 64}, "prefill_tps": [1837.733844453059, 2376.84065883052, 2360.0125375666057], "seq_len": 128}, {"decode_tps": [302.2061045633122, 302.8410273881854, 302.58473554566905], "native": {"agg": "mean", "decode_ms": 211.54, "decode_ms_best": 211.33, "decode_ms_worst": 211.78, "decode_n": 64, "decode_tps": 302.544, "prefill_ms": 109.63, "prefill_ms_best": 108.69, "prefill_ms_worst": 110.2, "prefill_tps": 2335.148, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.56, "samples": {"decode_ms": [211.776, 211.332, 211.511], "prefill_ms": [110.196, 109.996, 108.695], "total_ms": [321.972, 321.328, 320.205], "ttft_ms": [110.197, 109.997, 108.695]}, "seq_len": 256, "threads": 16, "total_ms": 321.17, "total_ms_best": 320.21, "total_ms_worst": 321.97, "total_tps": 996.361, "total_tps_best": 999.359, "total_tps_worst": 993.874, "ttft_ms": 109.63, "warmup": 64}, "prefill_tps": [2323.133326073542, 2327.3573584493984, 2355.214131284788], "seq_len": 256}, {"decode_tps": [268.08079284894484, 270.96369933190516, 270.8536120868425], "native": {"agg": "mean", "decode_ms": 237.07, "decode_ms_best": 236.19, "decode_ms_worst": 238.73, "decode_n": 64, "decode_tps": 269.959, "prefill_ms": 229.64, "prefill_ms_best": 227.29, "prefill_ms_worst": 231.17, "prefill_tps": 2229.567, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.56, "samples": {"decode_ms": [238.734, 236.194, 236.29], "prefill_ms": [230.467, 231.169, 227.287], "total_ms": [469.201, 467.364, 463.577], "ttft_ms": [230.467, 231.17, 227.288]}, "seq_len": 512, "threads": 16, "total_ms": 466.71, "total_ms_best": 463.58, "total_ms_worst": 469.2, "total_tps": 1234.161, "total_tps_best": 1242.511, "total_tps_worst": 1227.62, "ttft_ms": 229.64, "warmup": 64}, "prefill_tps": [2221.576190951416, 2214.8298431017997, 2252.658533044125], "seq_len": 512}, {"decode_tps": [213.64097633926187, 214.61026440655235, 218.60088601671612], "native": {"agg": "mean", "decode_ms": 296.85, "decode_ms_best": 292.77, "decode_ms_worst": 299.57, "decode_n": 64, "decode_tps": 215.596, "prefill_ms": 511.41, "prefill_ms_best": 509.69, "prefill_ms_worst": 512.48, "prefill_tps": 2002.301, "repeats": 3, "repeats_requested": 3, "rss_mb": 398.56, "samples": {"decode_ms": [299.568, 298.215, 292.771], "prefill_ms": [512.485, 509.688, 512.063], "total_ms": [812.052, 807.902, 804.834], "ttft_ms": [512.485, 509.688, 512.063]}, "seq_len": 1024, "threads": 16, "total_ms": 808.26, "total_ms_best": 804.83, "total_ms_worst": 812.05, "total_tps": 1346.097, "total_tps_best": 1351.832, "total_tps_worst": 1339.815, "ttft_ms": 511.41, "warmup": 64}, "prefill_tps": [1998.1072616759513, 2009.0722167286654, 1999.7539365273415], "seq_len": 1024}], "stderr": ["[bench] backend: cpu_x86", "[bench] warmup phase (discarded): 64-token prefill + 4 decode steps", "[bench] measuring 3 repeat(s) per seq_len; reporting the MEAN (with best/worst)"]}
{"cycle": 2, "elapsed_seconds": 11.69776810402982, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 2.92626953125, "load_per_core": 0.0914459228515625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          17.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.5\u00b0C  ", "Tccd1:        +45.2\u00b0C  ", "Tccd2:        +35.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +32.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.64501953125, "load_per_core": 0.0201568603515625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.07 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:           8.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +26.8\u00b0C  ", "Tccd2:        +27.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [231.558, 229.202, 227.984], "native": {"decode": {"avg_ns": 278779909, "avg_ts": 229.58134, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [276389092, 279229586, 280721050], "samples_ts": [231.558, 229.202, 227.984], "split_mode": "layer", "stddev_ns": 2200772, "stddev_ts": 1.816586, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:19Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 94540856, "avg_ts": 1353.923946, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 128, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [94763473, 94145176, 94713920], "samples_ts": [1350.73, 1359.6, 1351.44], "split_mode": "layer", "stddev_ns": 343701, "stddev_ts": 4.930297, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:13Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1350.73, 1359.6, 1351.44], "seq_len": 128}, {"decode_tps": [213.511, 217.162, 217.221], "native": {"decode": {"avg_ns": 296364181, "avg_ts": 215.964542, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [299750309, 294710749, 294631486], "samples_ts": [213.511, 217.162, 217.221], "split_mode": "layer", "stddev_ns": 2932790, "stddev_ts": 2.124997, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:20Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 182069096, "avg_ts": 1406.139227, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 256, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [180150527, 183231720, 182825043], "samples_ts": [1421.03, 1397.14, 1400.25], "split_mode": "layer", "stddev_ns": 1674034, "stddev_ts": 12.992395, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:13Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1421.03, 1397.14, 1400.25], "seq_len": 256}, {"decode_tps": [180.324, 182.853, 184.12], "native": {"decode": {"avg_ns": 350841590, "avg_ts": 182.432208, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [354917408, 350008485, 347598879], "samples_ts": [180.324, 182.853, 184.12], "split_mode": "layer", "stddev_ns": 3729807, "stddev_ts": 1.932929, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:21Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 372702583, "avg_ts": 1373.751499, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 512, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [373001226, 373033667, 372072856], "samples_ts": [1372.65, 1372.53, 1376.07], "split_mode": "layer", "stddev_ns": 545600, "stddev_ts": 2.012734, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:14Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1372.65, 1372.53, 1376.07], "seq_len": 512}, {"decode_tps": [138.519, 142.797, 144.157], "native": {"decode": {"avg_ns": 451393730, "avg_ts": 141.824164, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [462031276, 448188215, 443961700], "samples_ts": [138.519, 142.797, 144.157], "split_mode": "layer", "stddev_ns": 9451684, "stddev_ts": 2.942155, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:22Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 773312918, "avg_ts": 1324.17314, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 1024, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [773096489, 773052237, 773790029], "samples_ts": [1324.54, 1324.62, 1323.36], "split_mode": "layer", "stddev_ns": 414715, "stddev_ts": 0.708319, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:11:16Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1324.54, 1324.62, 1323.36], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 3, "elapsed_seconds": 11.673059403896332, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 3.01611328125, "load_per_core": 0.0942535400390625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.38 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.1\u00b0C  ", "Tccd1:        +38.0\u00b0C  ", "Tccd2:        +42.0\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.74365234375, "load_per_core": 0.0232391357421875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.1\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [226.877, 229.639, 228.102], "native": {"decode": {"avg_ns": 280455096, "avg_ts": 228.206115, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [282090677, 278698040, 280576571], "samples_ts": [226.877, 229.639, 228.102], "split_mode": "layer", "stddev_ns": 1699577, "stddev_ts": 1.383864, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:12:55Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 94087187, "avg_ts": 1360.445445, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 128, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [94296157, 93850107, 94115298], "samples_ts": [1357.43, 1363.88, 1360.03], "split_mode": "layer", "stddev_ns": 224559, "stddev_ts": 3.245405, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:13:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1357.43, 1363.88, 1360.03], "seq_len": 128}, {"decode_tps": [203.572, 218.17, 217.296], "native": {"decode": {"avg_ns": 300754321, "avg_ts": 213.012759, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [314384400, 293349756, 294528808], "samples_ts": [203.572, 218.17, 217.296], "split_mode": "layer", "stddev_ns": 11818719, "stddev_ts": 8.187209, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:12:56Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 181938858, "avg_ts": 1407.095193, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 256, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [181421592, 181286907, 183108076], "samples_ts": [1411.08, 1412.13, 1398.08], "split_mode": "layer", "stddev_ns": 1014898, "stddev_ts": 7.823731, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:13:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1411.08, 1412.13, 1398.08], "seq_len": 256}, {"decode_tps": [181.911, 184.821, 184.371], "native": {"decode": {"avg_ns": 348409721, "avg_ts": 183.700718, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [351820646, 346281906, 347126612], "samples_ts": [181.911, 184.821, 184.371], "split_mode": "layer", "stddev_ns": 2984046, "stddev_ts": 1.56628, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:12:57Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 370939846, "avg_ts": 1380.28487, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 512, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [371232700, 371796383, 369790457], "samples_ts": [1379.19, 1377.1, 1384.57], "split_mode": "layer", "stddev_ns": 1034890, "stddev_ts": 3.85375, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:13:02Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1379.19, 1377.1, 1384.57], "seq_len": 512}, {"decode_tps": [143.47, 144.7, 145.761], "native": {"decode": {"avg_ns": 442485355, "avg_ts": 144.64362, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 0, "n_prompt": 0, "n_threads": 15, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [446087262, 442293976, 439074827], "samples_ts": [143.47, 144.7, 145.761], "split_mode": "layer", "stddev_ns": 3510132, "stddev_ts": 1.146718, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:12:59Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 772984240, "avg_ts": 1324.735946, "backends": "CPU", "build_commit": "d81235049", "build_number": 11429, "cpu_info": "AMD Ryzen 9 9950X 16-Core Processor", "cpu_mask": "0x0", "cpu_strict": false, "devices": "auto", "embeddings": false, "fit_min_ctx": 0, "fit_target": 0, "flash_attn": -1, "gpu_info": "", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 0, "n_prompt": 1024, "n_threads": 16, "n_ubatch": 512, "no_host": false, "no_kv_offload": true, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [772910491, 773081057, 772961172], "samples_ts": [1324.86, 1324.57, 1324.78], "split_mode": "layer", "stddev_ns": 87591, "stddev_ts": 0.150108, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T13:13:03Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [1324.86, 1324.57, 1324.78], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 3, "elapsed_seconds": 5.648994204122573, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 2.10498046875, "load_per_core": 0.0657806396484375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.25 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.0\u00b0C  ", "Tccd1:        +36.9\u00b0C  ", "Tccd2:        +36.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +32.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.89501953125, "load_per_core": 0.0279693603515625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.9\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 1, "rows": [{"decode_tps": [324.55513126733706, 323.1392982828176, 323.2894703609224], "native": {"agg": "mean", "decode_ms": 197.74, "decode_ms_best": 197.19, "decode_ms_worst": 198.06, "decode_n": 64, "decode_tps": 323.66, "prefill_ms": 54.89, "prefill_ms_best": 54.13, "prefill_ms_worst": 55.57, "prefill_tps": 2331.915, "repeats": 3, "repeats_requested": 3, "rss_mb": 399.07, "samples": {"decode_ms": [197.193, 198.057, 197.965], "prefill_ms": [54.132, 54.971, 55.569], "total_ms": [251.325, 253.028, 253.534], "ttft_ms": [54.132, 54.972, 55.569]}, "seq_len": 128, "threads": 16, "total_ms": 252.63, "total_ms_best": 251.33, "total_ms_worst": 253.53, "total_tps": 760.008, "total_tps_best": 763.951, "total_tps_worst": 757.295, "ttft_ms": 54.89, "warmup": 64}, "prefill_tps": [2364.5902608438632, 2328.5004820723657, 2303.4425668988106], "seq_len": 128}, {"decode_tps": [306.7175945672646, 308.19163741253857, 302.3688711247177], "native": {"agg": "mean", "decode_ms": 209.33, "decode_ms_best": 207.66, "decode_ms_worst": 211.66, "decode_n": 64, "decode_tps": 305.739, "prefill_ms": 118.58, "prefill_ms_best": 109.95, "prefill_ms_worst": 127.25, "prefill_tps": 2158.857, "repeats": 3, "repeats_requested": 3, "rss_mb": 399.07, "samples": {"decode_ms": [208.661, 207.663, 211.662], "prefill_ms": [127.255, 109.947, 118.542], "total_ms": [335.916, 317.611, 330.204], "ttft_ms": [127.255, 109.948, 118.542]}, "seq_len": 256, "threads": 16, "total_ms": 327.91, "total_ms_best": 317.61, "total_ms_worst": 335.92, "total_tps": 975.877, "total_tps_best": 1007.522, "total_tps_worst": 952.619, "ttft_ms": 118.58, "warmup": 64}, "prefill_tps": [2011.7087737220543, 2328.3945901206944, 2159.5721347707986], "seq_len": 256}, {"decode_tps": [275.22975233622753, 274.48844360763593, 275.26763325749135], "native": {"agg": "mean", "decode_ms": 232.73, "decode_ms_best": 232.5, "decode_ms_worst": 233.16, "decode_n": 64, "decode_tps": 274.995, "prefill_ms": 233.46, "prefill_ms_best": 231.37, "prefill_ms_worst": 235.46, "prefill_tps": 2193.049, "repeats": 3, "repeats_requested": 3, "rss_mb": 399.07, "samples": {"decode_ms": [232.533, 233.161, 232.501], "prefill_ms": [235.462, 231.366, 233.567], "total_ms": [467.995, 464.527, 466.068], "ttft_ms": [235.462, 231.366, 233.568]}, "seq_len": 512, "threads": 16, "total_ms": 466.2, "total_ms_best": 464.53, "total_ms_worst": 467.99, "total_tps": 1235.531, "total_tps_best": 1239.972, "total_tps_worst": 1230.783, "ttft_ms": 233.47, "warmup": 64}, "prefill_tps": [2174.448530973151, 2212.943993499477, 2192.0904922356326], "seq_len": 512}, {"decode_tps": [214.19582853623928, 215.40480958551402, 215.45339473755084], "native": {"agg": "mean", "decode_ms": 297.65, "decode_ms_best": 297.05, "decode_ms_worst": 298.79, "decode_n": 64, "decode_tps": 215.016, "prefill_ms": 513.05, "prefill_ms_best": 512.01, "prefill_ms_worst": 513.8, "prefill_tps": 1995.891, "repeats": 3, "repeats_requested": 3, "rss_mb": 399.07, "samples": {"decode_ms": [298.792, 297.115, 297.048], "prefill_ms": [512.007, 513.801, 513.355], "total_ms": [810.798, 810.916, 810.402], "ttft_ms": [512.007, 513.801, 513.355]}, "seq_len": 1024, "threads": 16, "total_ms": 810.71, "total_ms_best": 810.4, "total_ms_worst": 810.92, "total_tps": 1342.041, "total_tps_best": 1342.543, "total_tps_worst": 1341.693, "ttft_ms": 513.05, "warmup": 64}, "prefill_tps": [1999.9726566238355, 1992.9895037183655, 1994.7210020356283], "seq_len": 1024}], "stderr": ["[bench] backend: cpu_x86", "[bench] warmup phase (discarded): 64-token prefill + 4 decode steps", "[bench] measuring 3 repeat(s) per seq_len; reporting the MEAN (with best/worst)"]}
{"kind": "summary", "summary": {"geist": {"1024": {"decode": {"mad_pct": 0.8648790221490723, "mad_tps": 1.8933597470478674, "median_tps": 218.91613723537904, "samples": 12, "vs_llama_pct": 51.57405722234814}, "prefill": {"mad_pct": 0.19999489297601483, "mad_tps": 3.9962489454698016, "median_tps": 1998.1754963858339, "samples": 12, "vs_llama_pct": 50.84155828637253}}, "128": {"decode": {"mad_pct": 0.9475583430720872, "mad_tps": 3.0338969212534153, "median_tps": 320.1804874006166, "samples": 12, "vs_llama_pct": 39.788421739959624}, "prefill": {"mad_pct": 3.1674392785224623, "mad_tps": 73.35699174595197, "median_tps": 2315.971524485588, "samples": 12, "vs_llama_pct": 70.31519173458065}}, "256": {"decode": {"mad_pct": 0.33593674486389563, "mad_tps": 1.0161307270926159, "median_tps": 302.4768033351934, "samples": 12, "vs_llama_pct": 39.49244069857332}, "prefill": {"mad_pct": 0.9905516315358405, "mad_tps": 23.032755674982127, "median_tps": 2325.24534226147, "samples": 12, "vs_llama_pct": 65.84433262805005}}, "512": {"decode": {"mad_pct": 0.6786829662400674, "mad_tps": 1.835143197606527, "median_tps": 270.39770981336085, "samples": 12, "vs_llama_pct": 46.7204446217119}, "prefill": {"mad_pct": 0.6279557506135809, "mad_tps": 13.859922846636664, "median_tps": 2207.1496014001013, "samples": 12, "vs_llama_pct": 60.76667478094393}}}, "llama.cpp": {"1024": {"decode": {"mad_pct": 0.9264099537141272, "mad_tps": 1.338000000000008, "median_tps": 144.42849999999999, "samples": 12}, "prefill": {"mad_pct": 0.012078343153275547, "mad_tps": 0.15999999999996817, "median_tps": 1324.685, "samples": 12}}, "128": {"decode": {"mad_pct": 0.4381206436247549, "mad_tps": 1.003499999999974, "median_tps": 229.04649999999998, "samples": 12}, "prefill": {"mad_pct": 0.2750374131775285, "mad_tps": 3.740000000000009, "median_tps": 1359.815, "samples": 12}}, "256": {"decode": {"mad_pct": 1.0297868023113663, "mad_tps": 2.23299999999999, "median_tps": 216.841, "samples": 12}, "prefill": {"mad_pct": 0.49712388512657496, "mad_tps": 6.969999999999914, "median_tps": 1402.065, "samples": 12}}, "512": {"decode": {"mad_pct": 0.5339280336635085, "mad_tps": 0.9839999999999947, "median_tps": 184.2945, "samples": 12}, "prefill": {"mad_pct": 0.37584948539212665, "mad_tps": 5.159999999999968, "median_tps": 1372.89, "samples": 12}}}}}