{"created_utc": "2026-10-12T00:52:40.404486+00:00", "engines": [{"binary": "/home/germar/gitea-runner-geisten/work/52e29b5a49e8b372/hostexecutor/bin/linux/release/tests/bench_perf_sweep", "binary_sha256": "34144480ea334b2185a0450db81b66c19619d4dabe903eb1b53e2c3359f84de3", "commit": "4d6e2696a51b5386017939d6c1f853cb9cb272fc", "kind": "geist", "label": "geist", "linked_libraries": ["linux-vdso.so.1 (0x00007ea89ee4e000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x00007ea89eb77000)", "\tlibmvec.so.1 => /usr/lib/x86_64-linux-gnu/libmvec.so.1 (0x00007ea89ea7c000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007ea89ea1d000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x00007ea89e600000)", "\t/lib64/ld-linux-x86-64.so.2 (0x00007ea89ee50000)"]}, {"binary": "/home/germar/llama.cpp-cross-engine/build-cpu/bin/llama-bench", "binary_sha256": "c7c9a15df0f96986765c2d61f67ba0081d5db8a7f06aad6ce0ab66b5bae018af", "commit": "d81235049384534c167caea52b85a694f6103d14", "kind": "llama", "label": "llama.cpp", "linked_libraries": ["linux-vdso.so.1 (0x000073abadd62000)", "\tlibllama-bench-impl.so => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libllama-bench-impl.so (0x000073abadcef000)", "\tlibstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x000073abada00000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x000073abadca4000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x000073abad600000)", "\tlibllama-common.so.0 => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libllama-common.so.0 (0x000073abad000000)", "\tlibllama.so.0 => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libllama.so.0 (0x000073abaca00000)", "\tlibggml.so.0 => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libggml.so.0 (0x000073abad9f2000)", "\tlibggml-base.so.0 => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libggml-base.so.0 (0x000073abad917000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x000073abaceda000)", "\t/lib64/ld-linux-x86-64.so.2 (0x000073abadd64000)", "\tlibssl.so.3 => /usr/lib/x86_64-linux-gnu/libssl.so.3 (0x000073abac8ef000)", "\tlibcrypto.so.3 => /usr/lib/x86_64-linux-gnu/libcrypto.so.3 (0x000073abac200000)", "\tlibggml-cpu.so.0 => /home/germar/llama.cpp-cross-engine/build-cpu/bin/libggml-cpu.so.0 (0x000073ababe00000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x000073abad8b8000)", "\tlibz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1 (0x000073abad89a000)", "\tlibzstd.so.1 => /usr/lib/x86_64-linux-gnu/libzstd.so.1 (0x000073abad53d000)"]}], "host_profile": "amd_9950x", "kind": "metadata", "model": {"file": "gemma4-e2b-Q4_K_M.gguf", "key": "gemma4-e2b-q4km", "path": "/home/germar/geist-models/gemma4-e2b-Q4_K_M.gguf", "sha256": "740185b21d22ceb83a11c3aa62ad5842ef32c70f6096d756bbee85a1e4ec34b8"}, "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": "4d6e2696a51b5386017939d6c1f853cb9cb272fc", "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": "63813234688", "os": "Linux 7.0.0-38-generic", "performance_cores": "16", "product": "Ubuntu 26.04.1 LTS"}}
{"cycle": 0, "elapsed_seconds": 26.279165255022235, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 5.41552734375, "load_per_core": 0.1692352294921875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.31 V  ", "vddnb:         1.19 V  ", "edge:         +41.0\u00b0C  ", "PPT:          17.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.5\u00b0C  ", "Tccd1:        +38.4\u00b0C  ", "Tccd2:        +36.5\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 0.32958984375, "load_per_core": 0.0102996826171875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.33 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          11.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.5\u00b0C  ", "Tccd1:        +25.6\u00b0C  ", "Tccd2:        +26.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "position": 0, "rows": [{"decode_tps": [51.143128840729105, 51.07647664887237, 52.00131953348316], "native": {"agg": "mean", "decode_ms": 1245.05, "decode_ms_best": 1230.74, "decode_ms_worst": 1253.02, "decode_n": 64, "decode_tps": 51.404, "prefill_ms": 242.12, "prefill_ms_best": 241.01, "prefill_ms_worst": 243.15, "prefill_tps": 528.663, "repeats": 3, "repeats_requested": 3, "rss_mb": 3319.07, "samples": {"decode_ms": [1251.39, 1253.023, 1230.738], "prefill_ms": [243.151, 242.202, 241.007], "total_ms": [1494.541, 1495.226, 1471.745], "ttft_ms": [243.152, 242.203, 241.008]}, "seq_len": 128, "threads": 16, "total_ms": 1487.17, "total_ms_best": 1471.74, "total_ms_worst": 1495.23, "total_tps": 129.104, "total_tps_best": 130.457, "total_tps_worst": 128.409, "ttft_ms": 242.12, "warmup": 64}, "prefill_tps": [526.4218530871763, 528.4844881545157, 531.1049056666403], "seq_len": 128}, {"decode_tps": [51.61943891283009, 51.47338544477428, 51.56046955475118], "native": {"agg": "mean", "decode_ms": 1241.49, "decode_ms_best": 1239.84, "decode_ms_worst": 1243.36, "decode_n": 64, "decode_tps": 51.551, "prefill_ms": 487.16, "prefill_ms_best": 483.44, "prefill_ms_worst": 493.03, "prefill_tps": 525.494, "repeats": 3, "repeats_requested": 3, "rss_mb": 3336.07, "samples": {"decode_ms": [1239.843, 1243.361, 1241.261], "prefill_ms": [485.008, 483.443, 493.031], "total_ms": [1724.85, 1726.804, 1734.292], "ttft_ms": [485.008, 483.444, 493.032]}, "seq_len": 256, "threads": 16, "total_ms": 1728.65, "total_ms_best": 1724.85, "total_ms_worst": 1734.29, "total_tps": 185.116, "total_tps_best": 185.523, "total_tps_worst": 184.513, "ttft_ms": 487.16, "warmup": 64}, "prefill_tps": [527.8263451324515, 529.5350227431155, 519.237127077202], "seq_len": 256}, {"decode_tps": [51.602790743426894, 51.78764476265075, 51.79472778228734], "native": {"agg": "mean", "decode_ms": 1237.24, "decode_ms_best": 1235.65, "decode_ms_worst": 1240.24, "decode_n": 64, "decode_tps": 51.728, "prefill_ms": 972.62, "prefill_ms_best": 963.73, "prefill_ms_worst": 977.71, "prefill_tps": 526.413, "repeats": 3, "repeats_requested": 3, "rss_mb": 3361.57, "samples": {"decode_ms": [1240.243, 1235.816, 1235.647], "prefill_ms": [976.426, 977.708, 963.729], "total_ms": [2216.67, 2213.524, 2199.376], "ttft_ms": [976.427, 977.708, 963.729]}, "seq_len": 512, "threads": 16, "total_ms": 2209.86, "total_ms_best": 2199.38, "total_ms_worst": 2216.67, "total_tps": 260.65, "total_tps_best": 261.892, "total_tps_worst": 259.849, "ttft_ms": 972.62, "warmup": 64}, "prefill_tps": [524.3612931241076, 523.6737348983542, 531.2696826597519], "seq_len": 512}, {"decode_tps": [51.35035383603191, 51.320459382262044, 51.49347159330456], "native": {"agg": "mean", "decode_ms": 1245.43, "decode_ms_best": 1242.88, "decode_ms_worst": 1247.07, "decode_n": 64, "decode_tps": 51.388, "prefill_ms": 1949.2, "prefill_ms_best": 1943.55, "prefill_ms_worst": 1959.18, "prefill_tps": 525.344, "repeats": 3, "repeats_requested": 3, "rss_mb": 3410.82, "samples": {"decode_ms": [1246.34, 1247.066, 1242.876], "prefill_ms": [1943.548, 1944.868, 1959.181], "total_ms": [3189.888, 3191.934, 3202.057], "ttft_ms": [1943.549, 1944.868, 1959.181]}, "seq_len": 1024, "threads": 16, "total_ms": 3194.63, "total_ms_best": 3189.89, "total_ms_worst": 3202.06, "total_tps": 340.572, "total_tps_best": 341.078, "total_tps_worst": 339.782, "ttft_ms": 1949.2, "warmup": 64}, "prefill_tps": [526.8714742316629, 526.5138816618917, 522.6673798898621], "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": 34.956043017999036, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 7.6376953125, "load_per_core": 0.238677978515625, "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:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +48.2\u00b0C  ", "Tccd1:        +40.2\u00b0C  ", "Tccd2:        +36.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.40478515625, "load_per_core": 0.0438995361328125, "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:         +35.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +28.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [44.4436, 44.6174, 44.4651], "native": {"decode": {"avg_ns": 1437924085, "avg_ts": 44.508743, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1440025932, 1434416357, 1439329968], "samples_ts": [44.4436, 44.6174, 44.4651], "split_mode": "layer", "stddev_ns": 3058118, "stddev_ts": 0.094754, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:20Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 218744091, "avg_ts": 585.168137, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [218176970, 218066424, 219988879], "samples_ts": [586.68, 586.977, 581.848], "split_mode": "layer", "stddev_ns": 1079434, "stddev_ts": 2.879517, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:07Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [586.68, 586.977, 581.848], "seq_len": 128}, {"decode_tps": [43.3565, 43.5406, 43.7497], "native": {"decode": {"avg_ns": 1469632189, "avg_ts": 43.548904, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1476135475, 1469892691, 1462868403], "samples_ts": [43.3565, 43.5406, 43.7497], "split_mode": "layer", "stddev_ns": 6637592, "stddev_ts": 0.196736, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:25Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 412733620, "avg_ts": 620.25719, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [412582234, 411820830, 413797797], "samples_ts": [620.482, 621.63, 618.66], "split_mode": "layer", "stddev_ns": 997346, "stddev_ts": 1.497703, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:08Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [620.482, 621.63, 618.66], "seq_len": 256}, {"decode_tps": [43.3182, 43.2836, 43.3827], "native": {"decode": {"avg_ns": 1477100119, "avg_ts": 43.328178, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1477437415, 1478619320, 1475243623], "samples_ts": [43.3182, 43.2836, 43.3827], "split_mode": "layer", "stddev_ns": 1713369, "stddev_ts": 0.050263, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:30Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 816261601, "avg_ts": 627.252323, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [814037848, 816969757, 817777200], "samples_ts": [628.963, 626.706, 626.087], "split_mode": "layer", "stddev_ns": 1968104, "stddev_ts": 1.513778, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:10Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [628.963, 626.706, 626.087], "seq_len": 512}, {"decode_tps": [42.6842, 42.7266, 42.6407], "native": {"decode": {"avg_ns": 1499398056, "avg_ts": 42.683824, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1499383913, 1497896499, 1500913756], "samples_ts": [42.6842, 42.7266, 42.6407], "split_mode": "layer", "stddev_ns": 1508678, "stddev_ts": 0.042947, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:35Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1691732127, "avg_ts": 605.315066, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1702955362, 1692054583, 1680186438], "samples_ts": [601.308, 605.181, 609.456], "split_mode": "layer", "stddev_ns": 11388035, "stddev_ts": 4.075932, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:56:13Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [601.308, 605.181, 609.456], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 1, "elapsed_seconds": 34.97787962300936, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 6.69091796875, "load_per_core": 0.2090911865234375, "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:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.2\u00b0C  ", "Tccd1:        +50.2\u00b0C  ", "Tccd2:        +35.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +30.0\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:    +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)"]}, "machine_before": {"load_1m": 1.259765625, "load_per_core": 0.03936767578125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.34 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          11.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.6\u00b0C  ", "Tccd1:        +26.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +31.9\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:    +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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [44.4341, 44.4454, 44.5046], "native": {"decode": {"avg_ns": 1439452906, "avg_ts": 44.461358, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1440334631, 1439969409, 1438054679], "samples_ts": [44.4341, 44.4454, 44.5046], "split_mode": "layer", "stddev_ns": 1225179, "stddev_ts": 0.037842, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:32Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 216116887, "avg_ts": 592.275473, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [216638351, 215413227, 216299085], "samples_ts": [590.846, 594.207, 591.773], "split_mode": "layer", "stddev_ns": 632899, "stddev_ts": 1.735562, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:54Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [590.846, 594.207, 591.773], "seq_len": 128}, {"decode_tps": [43.5589, 43.3085, 43.3541], "native": {"decode": {"avg_ns": 1474419362, "avg_ts": 43.407191, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1469273820, 1477769846, 1476214422], "samples_ts": [43.5589, 43.3085, 43.3541], "split_mode": "layer", "stddev_ns": 4523852, "stddev_ts": 0.133379, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:37Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 411980014, "avg_ts": 621.416653, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [415714161, 409242267, 410983614], "samples_ts": [615.808, 625.546, 622.896], "split_mode": "layer", "stddev_ns": 3349024, "stddev_ts": 5.034989, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:55Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [615.808, 625.546, 622.896], "seq_len": 256}, {"decode_tps": [43.0701, 43.0256, 43.1267], "native": {"decode": {"avg_ns": 1485811920, "avg_ts": 43.074132, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1485950269, 1487486271, 1483999222], "samples_ts": [43.0701, 43.0256, 43.1267], "split_mode": "layer", "stddev_ns": 1748486, "stddev_ts": 0.050672, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:42Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 825058083, "avg_ts": 620.694048, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [814162182, 819124817, 841887251], "samples_ts": [628.867, 625.057, 608.157], "split_mode": "layer", "stddev_ns": 14784228, "stddev_ts": 11.022875, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:57Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [628.867, 625.057, 608.157], "seq_len": 512}, {"decode_tps": [42.6097, 43.0529, 42.7076], "native": {"decode": {"avg_ns": 1495703352, "avg_ts": 42.790076, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1502005187, 1486543995, 1498560875], "samples_ts": [42.6097, 43.0529, 42.7076], "split_mode": "layer", "stddev_ns": 8117123, "stddev_ts": 0.232803, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:58:47Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1680972572, "avg_ts": 609.17846, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1686200095, 1672860224, 1683857398], "samples_ts": [607.283, 612.125, 608.128], "split_mode": "layer", "stddev_ns": 7122596, "stddev_ts": 2.586733, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T00:59:00Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [607.283, 612.125, 608.128], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 1, "elapsed_seconds": 26.306181635998655, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 6.18212890625, "load_per_core": 0.1931915283203125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.33 V  ", "vddnb:         1.19 V  ", "edge:         +41.0\u00b0C  ", "PPT:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.8\u00b0C  ", "Tccd1:        +38.9\u00b0C  ", "Tccd2:        +37.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.25830078125, "load_per_core": 0.0393218994140625, "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:         +35.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.8\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "position": 1, "rows": [{"decode_tps": [51.36754036966971, 51.545602958717616, 51.58765023895561], "native": {"agg": "mean", "decode_ms": 1242.72, "decode_ms_best": 1240.61, "decode_ms_worst": 1245.92, "decode_n": 64, "decode_tps": 51.5, "prefill_ms": 241.98, "prefill_ms_best": 241.37, "prefill_ms_worst": 242.31, "prefill_tps": 528.975, "repeats": 3, "repeats_requested": 3, "rss_mb": 3319.18, "samples": {"decode_ms": [1245.923, 1241.619, 1240.607], "prefill_ms": [242.259, 241.368, 242.306], "total_ms": [1488.181, 1482.986, 1482.912], "ttft_ms": [242.259, 241.368, 242.306]}, "seq_len": 128, "threads": 16, "total_ms": 1484.69, "total_ms_best": 1482.91, "total_ms_worst": 1488.18, "total_tps": 129.32, "total_tps_best": 129.475, "total_tps_worst": 129.017, "ttft_ms": 241.98, "warmup": 64}, "prefill_tps": [528.3601434828015, 530.3105631235293, 528.2576576725297], "seq_len": 128}, {"decode_tps": [51.192300677978025, 51.816324085198985, 51.42491199510178], "native": {"agg": "mean", "decode_ms": 1243.28, "decode_ms_best": 1235.13, "decode_ms_worst": 1250.19, "decode_n": 64, "decode_tps": 51.477, "prefill_ms": 483.45, "prefill_ms_best": 478.89, "prefill_ms_worst": 489.57, "prefill_tps": 529.524, "repeats": 3, "repeats_requested": 3, "rss_mb": 3336.18, "samples": {"decode_ms": [1250.188, 1235.132, 1244.533], "prefill_ms": [489.574, 481.899, 478.887], "total_ms": [1739.762, 1717.031, 1723.42], "ttft_ms": [489.574, 481.899, 478.888]}, "seq_len": 256, "threads": 16, "total_ms": 1726.74, "total_ms_best": 1717.03, "total_ms_worst": 1739.76, "total_tps": 185.321, "total_tps_best": 186.368, "total_tps_worst": 183.933, "ttft_ms": 483.45, "warmup": 64}, "prefill_tps": [522.903585566227, 531.2316481254371, 534.5728741853507], "seq_len": 256}, {"decode_tps": [51.34351539422995, 51.29425832910557, 51.43040822886531], "native": {"agg": "mean", "decode_ms": 1246.2, "decode_ms_best": 1244.4, "decode_ms_worst": 1247.7, "decode_n": 64, "decode_tps": 51.356, "prefill_ms": 971.88, "prefill_ms_best": 967.82, "prefill_ms_worst": 977.53, "prefill_tps": 526.813, "repeats": 3, "repeats_requested": 3, "rss_mb": 3361.68, "samples": {"decode_ms": [1246.506, 1247.703, 1244.4], "prefill_ms": [967.817, 977.529, 970.303], "total_ms": [2214.323, 2225.232, 2214.703], "ttft_ms": [967.817, 977.53, 970.303]}, "seq_len": 512, "threads": 16, "total_ms": 2218.09, "total_ms_best": 2214.32, "total_ms_worst": 2225.23, "total_tps": 259.683, "total_tps_best": 260.125, "total_tps_worst": 258.849, "ttft_ms": 971.88, "warmup": 64}, "prefill_tps": [529.025631911818, 523.7696272949447, 527.6702226005691], "seq_len": 512}, {"decode_tps": [51.51344903468626, 50.93927242800451, 51.125520341965824], "native": {"agg": "mean", "decode_ms": 1250.2, "decode_ms_best": 1242.39, "decode_ms_worst": 1256.4, "decode_n": 64, "decode_tps": 51.192, "prefill_ms": 1952.6, "prefill_ms_best": 1947.42, "prefill_ms_worst": 1960.62, "prefill_tps": 524.428, "repeats": 3, "repeats_requested": 3, "rss_mb": 3410.68, "samples": {"decode_ms": [1242.394, 1256.398, 1251.821], "prefill_ms": [1947.423, 1949.773, 1960.616], "total_ms": [3189.818, 3206.171, 3212.437], "ttft_ms": [1947.424, 1949.774, 1960.616]}, "seq_len": 1024, "threads": 16, "total_ms": 3202.81, "total_ms_best": 3189.82, "total_ms_worst": 3212.44, "total_tps": 339.702, "total_tps_best": 341.085, "total_tps_worst": 338.684, "ttft_ms": 1952.6, "warmup": 64}, "prefill_tps": [525.8231005795865, 525.1893425542359, 522.2848329300588], "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": 26.230965042981552, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 6.99609375, "load_per_core": 0.2186279296875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.30 V  ", "vddnb:         1.19 V  ", "edge:         +41.0\u00b0C  ", "PPT:          16.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.9\u00b0C  ", "Tccd1:        +39.8\u00b0C  ", "Tccd2:        +35.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.5673828125, "load_per_core": 0.048980712890625, "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:         +35.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "position": 0, "rows": [{"decode_tps": [51.83281648996265, 52.148094068643196, 51.68250789369554], "native": {"agg": "mean", "decode_ms": 1233.45, "decode_ms_best": 1227.27, "decode_ms_worst": 1238.33, "decode_n": 64, "decode_tps": 51.887, "prefill_ms": 241.48, "prefill_ms_best": 241.17, "prefill_ms_worst": 242.07, "prefill_tps": 530.069, "repeats": 3, "repeats_requested": 3, "rss_mb": 3318.73, "samples": {"decode_ms": [1234.739, 1227.274, 1238.33], "prefill_ms": [241.17, 241.194, 242.069], "total_ms": [1475.909, 1468.468, 1480.4], "ttft_ms": [241.171, 241.195, 242.07]}, "seq_len": 128, "threads": 16, "total_ms": 1474.93, "total_ms_best": 1468.47, "total_ms_worst": 1480.4, "total_tps": 130.176, "total_tps_best": 130.748, "total_tps_worst": 129.695, "ttft_ms": 241.48, "warmup": 64}, "prefill_tps": [530.7459468424763, 530.6931349867741, 528.7748534508756], "seq_len": 128}, {"decode_tps": [51.06274334588626, 51.373106820743175, 51.67240983971865], "native": {"agg": "mean", "decode_ms": 1245.91, "decode_ms_best": 1238.57, "decode_ms_worst": 1253.36, "decode_n": 64, "decode_tps": 51.368, "prefill_ms": 483.21, "prefill_ms_best": 481.05, "prefill_ms_worst": 484.39, "prefill_tps": 529.787, "repeats": 3, "repeats_requested": 3, "rss_mb": 3335.98, "samples": {"decode_ms": [1253.36, 1245.788, 1238.572], "prefill_ms": [484.391, 484.199, 481.049], "total_ms": [1737.752, 1729.987, 1719.622], "ttft_ms": [484.392, 484.2, 481.05]}, "seq_len": 256, "threads": 16, "total_ms": 1729.12, "total_ms_best": 1719.62, "total_ms_worst": 1737.75, "total_tps": 185.065, "total_tps_best": 186.087, "total_tps_worst": 184.146, "ttft_ms": 483.21, "warmup": 64}, "prefill_tps": [528.4986715277533, 528.7082377287024, 532.170319447707], "seq_len": 256}, {"decode_tps": [51.54182538595968, 51.45228082333296, 51.66202381134966], "native": {"agg": "mean", "decode_ms": 1241.47, "decode_ms_best": 1238.82, "decode_ms_worst": 1243.87, "decode_n": 64, "decode_tps": 51.552, "prefill_ms": 966.23, "prefill_ms_best": 963.22, "prefill_ms_worst": 969.16, "prefill_tps": 529.896, "repeats": 3, "repeats_requested": 3, "rss_mb": 3361.48, "samples": {"decode_ms": [1241.71, 1243.871, 1238.821], "prefill_ms": [969.156, 966.302, 963.223], "total_ms": [2210.866, 2210.174, 2202.045], "ttft_ms": [969.156, 966.303, 963.224]}, "seq_len": 512, "threads": 16, "total_ms": 2207.69, "total_ms_best": 2202.04, "total_ms_worst": 2210.87, "total_tps": 260.906, "total_tps_best": 261.575, "total_tps_worst": 260.531, "ttft_ms": 966.23, "warmup": 64}, "prefill_tps": [528.2947224182691, 529.8550556658271, 531.5487690804725], "seq_len": 512}, {"decode_tps": [51.647348675285855, 51.644598015394934, 51.70593220544383], "native": {"agg": "mean", "decode_ms": 1238.73, "decode_ms_best": 1237.77, "decode_ms_worst": 1239.24, "decode_n": 64, "decode_tps": 51.666, "prefill_ms": 1956.0, "prefill_ms_best": 1942.52, "prefill_ms_worst": 1969.91, "prefill_tps": 523.516, "repeats": 3, "repeats_requested": 3, "rss_mb": 3410.48, "samples": {"decode_ms": [1239.173, 1239.239, 1237.769], "prefill_ms": [1942.52, 1955.587, 1969.906], "total_ms": [3181.693, 3194.826, 3207.675], "ttft_ms": [1942.521, 1955.587, 1969.907]}, "seq_len": 1024, "threads": 16, "total_ms": 3194.73, "total_ms_best": 3181.69, "total_ms_worst": 3207.68, "total_tps": 340.561, "total_tps_best": 341.956, "total_tps_worst": 339.186, "ttft_ms": 1956.0, "warmup": 64}, "prefill_tps": [527.1502996108148, 523.6279439370378, 519.8217579925133], "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": 34.840464152017375, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 7.595703125, "load_per_core": 0.23736572265625, "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:         +41.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +48.2\u00b0C  ", "Tccd1:        +46.8\u00b0C  ", "Tccd2:        +34.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +27.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +27.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.40283203125, "load_per_core": 0.0438385009765625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.34 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          10.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.8\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [44.279, 44.3303, 44.3858], "native": {"decode": {"avg_ns": 1443664702, "avg_ts": 44.331666, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1445381966, 1443708821, 1441903321], "samples_ts": [44.279, 44.3303, 44.3858], "split_mode": "layer", "stddev_ns": 1740571, "stddev_ts": 0.053426, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:05:03Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 215382282, "avg_ts": 594.336838, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [213511792, 214695712, 217939342], "samples_ts": [599.499, 596.193, 587.319], "split_mode": "layer", "stddev_ns": 2292233, "stddev_ts": 6.298072, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:04:50Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [599.499, 596.193, 587.319], "seq_len": 128}, {"decode_tps": [43.6073, 43.642, 43.565], "native": {"decode": {"avg_ns": 1467730427, "avg_ts": 43.60476, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1467644145, 1466477393, 1469069743], "samples_ts": [43.6073, 43.642, 43.565], "split_mode": "layer", "stddev_ns": 1298327, "stddev_ts": 0.038569, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:05:08Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 410247072, "avg_ts": 624.015446, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [409750408, 411059855, 409930955], "samples_ts": [624.771, 622.78, 624.495], "split_mode": "layer", "stddev_ns": 710233, "stddev_ts": 1.078438, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:04:51Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [624.771, 622.78, 624.495], "seq_len": 256}, {"decode_tps": [43.1215, 43.158, 43.1855], "native": {"decode": {"avg_ns": 1483027352, "avg_ts": 43.154985, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1484178538, 1482922911, 1481980609], "samples_ts": [43.1215, 43.158, 43.1855], "split_mode": "layer", "stddev_ns": 1104024, "stddev_ts": 0.032084, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:05:13Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 812858300, "avg_ts": 629.877462, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [814555870, 812133616, 811885414], "samples_ts": [628.563, 630.438, 630.631], "split_mode": "layer", "stddev_ns": 1475367, "stddev_ts": 1.142092, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:04:53Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [628.563, 630.438, 630.631], "seq_len": 512}, {"decode_tps": [42.8341, 42.9149, 43.0256], "native": {"decode": {"avg_ns": 1490982773, "avg_ts": 42.924851, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1494137916, 1491323852, 1487486553], "samples_ts": [42.8341, 42.9149, 43.0256], "split_mode": "layer", "stddev_ns": 3339220, "stddev_ts": 0.096155, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:05:18Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1670227403, "avg_ts": 613.092224, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1667094439, 1669208706, 1674379066], "samples_ts": [614.242, 613.464, 611.57], "split_mode": "layer", "stddev_ns": 3748079, "stddev_ts": 1.374485, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:04:56Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [614.242, 613.464, 611.57], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 3, "elapsed_seconds": 34.896557916974416, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 7.640625, "load_per_core": 0.23876953125, "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:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.4\u00b0C  ", "Tccd1:        +40.2\u00b0C  ", "Tccd2:        +34.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.29638671875, "load_per_core": 0.0405120849609375, "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:         +35.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +28.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [44.3356, 44.125, 44.0595], "native": {"decode": {"avg_ns": 1448847581, "avg_ts": 44.173352, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1443536485, 1450424859, 1452581399], "samples_ts": [44.3356, 44.125, 44.0595], "split_mode": "layer", "stddev_ns": 4724242, "stddev_ts": 0.144245, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:15Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 216388192, "avg_ts": 591.540567, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [216679758, 215125601, 217359217], "samples_ts": [590.734, 595.001, 588.887], "split_mode": "layer", "stddev_ns": 1144997, "stddev_ts": 3.136034, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:37Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [590.734, 595.001, 588.887], "seq_len": 128}, {"decode_tps": [43.393, 43.4233, 43.6507], "native": {"decode": {"avg_ns": 1471646256, "avg_ts": 43.489016, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1474891584, 1473863378, 1466183807], "samples_ts": [43.393, 43.4233, 43.6507], "split_mode": "layer", "stddev_ns": 4758627, "stddev_ts": 0.140868, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:20Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 411102217, "avg_ts": 622.749167, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [415107173, 410299117, 407900363], "samples_ts": [616.708, 623.935, 627.604], "split_mode": "layer", "stddev_ns": 3670024, "stddev_ts": 5.543975, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [616.708, 623.935, 627.604], "seq_len": 256}, {"decode_tps": [43.2173, 42.777, 43.0605], "native": {"decode": {"avg_ns": 1487767753, "avg_ts": 43.01824, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1480890034, 1496132107, 1486281119], "samples_ts": [43.2173, 42.777, 43.0605], "split_mode": "layer", "stddev_ns": 7729116, "stddev_ts": 0.223162, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:25Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 809612309, "avg_ts": 632.405822, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [808754823, 812541466, 807540640], "samples_ts": [633.072, 630.122, 634.024], "split_mode": "layer", "stddev_ns": 2608667, "stddev_ts": 2.034566, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:40Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [633.072, 630.122, 634.024], "seq_len": 512}, {"decode_tps": [42.5623, 42.6786, 42.7132], "native": {"decode": {"avg_ns": 1500540339, "avg_ts": 42.6514, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1503676453, 1499580105, 1498364461], "samples_ts": [42.5623, 42.6786, 42.7132], "split_mode": "layer", "stddev_ns": 2783676, "stddev_ts": 0.079041, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:30Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1663389456, "avg_ts": 615.620438, "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/geist-models/gemma4-e2b-Q4_K_M.gguf", "model_n_params": 4647450147, "model_size": 3090917516, "model_type": "gemma4 E2B Q4_K - Medium", "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": [1662363965, 1655757938, 1672046465], "samples_ts": [615.99, 618.448, 612.423], "split_mode": "layer", "stddev_ns": 8192542, "stddev_ts": 3.029335, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:07:43Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [615.99, 618.448, 612.423], "seq_len": 1024}], "stderr": {"decode": [], "prefill": []}}
{"cycle": 3, "elapsed_seconds": 26.271209717990132, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 6.0380859375, "load_per_core": 0.188690185546875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.34 V  ", "vddnb:         1.19 V  ", "edge:         +41.0\u00b0C  ", "PPT:          21.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.8\u00b0C  ", "Tccd1:        +40.6\u00b0C  ", "Tccd2:        +36.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "machine_before": {"load_1m": 1.27294921875, "load_per_core": 0.0397796630859375, "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:           8.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +28.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +28.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +28.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +30.9\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:    +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)"]}, "position": 1, "rows": [{"decode_tps": [51.769547857711395, 51.76854284494028, 51.7479486789719], "native": {"agg": "mean", "decode_ms": 1236.43, "decode_ms_best": 1236.25, "decode_ms_worst": 1236.76, "decode_n": 64, "decode_tps": 51.762, "prefill_ms": 246.33, "prefill_ms_best": 242.9, "prefill_ms_worst": 249.91, "prefill_tps": 519.629, "repeats": 3, "repeats_requested": 3, "rss_mb": 3318.75, "samples": {"decode_ms": [1236.248, 1236.272, 1236.764], "prefill_ms": [242.896, 249.907, 246.186], "total_ms": [1479.144, 1486.179, 1482.95], "ttft_ms": [242.896, 249.907, 246.187]}, "seq_len": 128, "threads": 16, "total_ms": 1482.76, "total_ms_best": 1479.14, "total_ms_worst": 1486.18, "total_tps": 129.488, "total_tps_best": 129.805, "total_tps_worst": 129.19, "ttft_ms": 246.33, "warmup": 64}, "prefill_tps": [526.9745076081945, 512.190534878975, 519.9320838715443], "seq_len": 128}, {"decode_tps": [51.77964202468109, 51.62680915665994, 51.576009741418844], "native": {"agg": "mean", "decode_ms": 1238.85, "decode_ms_best": 1236.01, "decode_ms_worst": 1240.89, "decode_n": 64, "decode_tps": 51.661, "prefill_ms": 483.3, "prefill_ms_best": 480.5, "prefill_ms_worst": 484.99, "prefill_tps": 529.69, "repeats": 3, "repeats_requested": 3, "rss_mb": 3335.75, "samples": {"decode_ms": [1236.007, 1239.666, 1240.887], "prefill_ms": [480.495, 484.422, 484.987], "total_ms": [1716.502, 1724.088, 1725.874], "ttft_ms": [480.496, 484.422, 484.987]}, "seq_len": 256, "threads": 16, "total_ms": 1722.15, "total_ms_best": 1716.5, "total_ms_worst": 1725.87, "total_tps": 185.814, "total_tps_best": 186.426, "total_tps_worst": 185.413, "ttft_ms": 483.3, "warmup": 64}, "prefill_tps": [532.7838999365238, 528.464850894468, 527.8492000816517], "seq_len": 256}, {"decode_tps": [51.46974262716042, 50.8999751862621, 51.54676540020023], "native": {"agg": "mean", "decode_ms": 1247.47, "decode_ms_best": 1241.59, "decode_ms_worst": 1257.37, "decode_n": 64, "decode_tps": 51.304, "prefill_ms": 969.44, "prefill_ms_best": 967.68, "prefill_ms_worst": 971.96, "prefill_tps": 528.141, "repeats": 3, "repeats_requested": 3, "rss_mb": 3361.25, "samples": {"decode_ms": [1243.449, 1257.368, 1241.591], "prefill_ms": [967.685, 968.672, 971.959], "total_ms": [2211.134, 2226.04, 2213.55], "ttft_ms": [967.685, 968.672, 971.96]}, "seq_len": 512, "threads": 16, "total_ms": 2216.91, "total_ms_best": 2211.13, "total_ms_worst": 2226.04, "total_tps": 259.821, "total_tps_best": 260.5, "total_tps_worst": 258.756, "ttft_ms": 969.44, "warmup": 64}, "prefill_tps": [529.0977952536208, 528.558686531664, 526.7711909658741], "seq_len": 512}, {"decode_tps": [51.03428148117433, 51.288914462111514, 51.441900392405245], "native": {"agg": "mean", "decode_ms": 1248.67, "decode_ms_best": 1244.12, "decode_ms_worst": 1254.06, "decode_n": 64, "decode_tps": 51.254, "prefill_ms": 1952.52, "prefill_ms_best": 1951.12, "prefill_ms_worst": 1954.46, "prefill_tps": 524.449, "repeats": 3, "repeats_requested": 3, "rss_mb": 3410.5, "samples": {"decode_ms": [1254.059, 1247.833, 1244.122], "prefill_ms": [1951.989, 1951.124, 1954.462], "total_ms": [3206.048, 3198.957, 3198.583], "ttft_ms": [1951.99, 1951.124, 1954.462]}, "seq_len": 1024, "threads": 16, "total_ms": 3201.2, "total_ms_best": 3198.58, "total_ms_worst": 3206.05, "total_tps": 339.873, "total_tps_best": 340.151, "total_tps_worst": 339.359, "ttft_ms": 1952.53, "warmup": 64}, "prefill_tps": [524.5931201456566, 524.8256902175361, 523.9293473088758], "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": 0.3558564060976242, "mad_tps": 0.1828964108220248, "median_tps": 51.39612711421857, "samples": 12, "vs_llama_pct": 20.336328187557527}, "prefill": {"mad_pct": 0.278075049681592, "mad_tps": 1.4590859391427102, "median_tps": 524.7094051815964, "samples": 12, "vs_llama_pct": -14.241799601764116}}, "128": {"decode": {"mad_pct": 0.2873460920917607, "mad_tps": 0.14860168749711278, "median_tps": 51.715228286333726, "samples": 12, "vs_llama_pct": 16.44964312352013}, "prefill": {"mad_pct": 0.3679547513363844, "mad_tps": 1.944355018176509, "median_tps": 528.4223158186586, "samples": 12, "vs_llama_pct": -10.556658741911907}}, "256": {"decode": {"mad_pct": 0.239971197684714, "mad_tps": 0.12374892230843315, "median_tps": 51.56823964808501, "samples": 12, "vs_llama_pct": 18.412251845498552}, "prefill": {"mad_pct": 0.3367138051150819, "mad_tps": 1.7798808060484816, "median_tps": 528.6034546282278, "samples": 12, "vs_llama_pct": -15.129864486715993}}, "512": {"decode": {"mad_pct": 0.24584281798739757, "mad_tps": 0.12662327082822955, "median_tps": 51.50578400656005, "samples": 12, "vs_llama_pct": 19.385670939483003}, "prefill": {"mad_pct": 0.2917968257316826, "mad_tps": 1.5419323499764914, "median_tps": 528.4267044749665, "samples": 12, "vs_llama_pct": -15.978040836207352}}}, "llama.cpp": {"1024": {"decode": {"mad_pct": 0.19948302989435918, "mad_tps": 0.08520000000000039, "median_tps": 42.7104, "samples": 12}, "prefill": {"mad_pct": 0.4996343042996758, "mad_tps": 3.0569999999999595, "median_tps": 611.8475000000001, "samples": 12}}, "128": {"decode": {"mad_pct": 0.17338456809791888, "mad_tps": 0.07700000000000173, "median_tps": 44.40995, "samples": 12}, "prefill": {"mad_pct": 0.6164627024831152, "mad_tps": 3.641999999999996, "median_tps": 590.79, "samples": 12}}, "256": {"decode": {"mad_pct": 0.26108071802938354, "mad_tps": 0.11370000000000147, "median_tps": 43.54975, "samples": 12}, "prefill": {"mad_pct": 0.3443110407521689, "mad_tps": 2.1444999999999936, "median_tps": 622.838, "samples": 12}}, "512": {"decode": {"mad_pct": 0.181723990464131, "mad_tps": 0.07840000000000202, "median_tps": 43.14235, "samples": 12}, "prefill": {"mad_pct": 0.3120453479404971, "mad_tps": 1.9624999999999773, "median_tps": 628.915, "samples": 12}}}}}