{"created_utc": "2026-10-12T01:26:23.532451+00:00", "engines": [{"binary": "/home/germar/gitea-runner-geisten/work/8b5de658f5231e12/hostexecutor/bin/linux/release/tests/bench_perf_sweep", "binary_sha256": "b1a3283699b67735d3ea8fc0d943c3765f26da65624707a46012948ae14d77cd", "commit": "4d6e2696a51b5386017939d6c1f853cb9cb272fc", "kind": "geist", "label": "geist", "linked_libraries": ["linux-vdso.so.1 (0x000074d8bd3a3000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x000074d8bceda000)", "\tlibmvec.so.1 => /usr/lib/x86_64-linux-gnu/libmvec.so.1 (0x000074d8bcddf000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x000074d8bd31f000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x000074d8bca00000)", "\t/lib64/ld-linux-x86-64.so.2 (0x000074d8bd3a5000)"]}, {"binary": "/home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/llama-bench", "binary_sha256": "efacee59b5b8a6fbbac816370d0ce127bcdd7a9bc14861b1a58314e8a96b93f3", "commit": "adfffbe41b2cabcd51fff326ab045662265062bb", "kind": "llama", "label": "llama.cpp", "linked_libraries": ["linux-vdso.so.1 (0x00007d04eb54e000)", "\tlibllama-bench-impl.so => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libllama-bench-impl.so (0x00007d04eb4dd000)", "\tlibstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007d04eb200000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007d04eb1d2000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x00007d04eae00000)", "\tlibllama-common.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libllama-common.so.0 (0x00007d04ea800000)", "\tlibllama.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libllama.so.0 (0x00007d04ea400000)", "\tlibggml.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libggml.so.0 (0x00007d04eb4b1000)", "\tlibggml-base.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libggml-base.so.0 (0x00007d04eb0f7000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x00007d04eacda000)", "\t/lib64/ld-linux-x86-64.so.2 (0x00007d04eb550000)", "\tlibssl.so.3 => /usr/lib/x86_64-linux-gnu/libssl.so.3 (0x00007d04ea2ef000)", "\tlibcrypto.so.3 => /usr/lib/x86_64-linux-gnu/libcrypto.so.3 (0x00007d04e9c00000)", "\tlibggml-cpu.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libggml-cpu.so.0 (0x00007d04e9a1c000)", "\tlibggml-vulkan.so.0 => /home/germar/llama.cpp-cross-engine-PrismML-Eng_llama.cpp/build-vulkan/bin/libggml-vulkan.so.0 (0x00007d04e6c00000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007d04eb098000)", "\tlibz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1 (0x00007d04eb07a000)", "\tlibzstd.so.1 => /usr/lib/x86_64-linux-gnu/libzstd.so.1 (0x00007d04ea22c000)", "\tlibvulkan.so.1 => /usr/lib/x86_64-linux-gnu/libvulkan.so.1 (0x00007d04e998c000)"]}], "host_profile": "nvidia_2080ti_vulkan", "kind": "metadata", "model": {"file": "Ternary-Bonsai-2-27B-PQ2_0.gguf", "key": "bonsai2-27b-pq2", "path": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "sha256": "3907dc1658db1f78a9826bf8d5bcb8dc65db0d466388937af57f2294fae62ec1"}, "protocol": {"host_gate": {"cooldown_seconds": 60, "max_load_per_core": 0.2, "settle_seconds": 30, "temp_gate": {"max_millic": 62000, "note": "same gate as the CPU protocol: GPU campaigns on passively cooled hosts throttle too. Hosts without the sensor path skip this gate.", "sensor": "/sys/class/thermal/thermal_zone0/temp"}, "timeout_seconds": 14400}, "host_profiles": {"apple_m1max_metal": {"decode_threads": 4, "expect_backend": "metal", "expect_llama_backend": "MTL", "note": "M1 Max 32-core GPU, unified memory. Same host as apple_m1max in the CPU protocol; the GPU cell and the CPU cell describe the same machine under two backends. Thread counts are dispatch-side only.", "prefill_threads": 4}, "nvidia_2080ti_vulkan": {"decode_threads": 4, "expect_backend": "vulkan", "expect_llama_backend": "Vulkan", "note": "RTX 2080 Ti 11 GB (Turing) in the 9950X box; same host as amd_9950x in the CPU protocol, different backend column. The host also enumerates a Raphael iGPU \u2014 runs must set GGML_VK_VISIBLE_DEVICES=0 so llama.cpp binds the discrete device; geist's vk_pick_device prefers discrete on its own. Thread counts are dispatch-side only, matching apple_m1max_metal.", "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-gpu.v2", "workload": {"affinity": "thread-count-only; operating-system scheduled for both engines", "aggregation": "median", "backend": "gpu", "cycles": 4, "decode_n": 64, "dispersion": "median_absolute_deviation", "gpu_offload_layers": 99, "kv_cache": "f16, resident on the GPU for both engines", "model_load_included": false, "repeats": 3, "seq_lens": [128, 256, 512, 1024], "thread_policy": "per host_profile; identical for both engines within a run. On a fully offloaded run CPU threads only drive dispatch and sampling; they are fixed anyway so a cell states its whole configuration.", "thread_polling": "active", "warmup": "one discarded engine-native warmup before each measured test"}}, "protocol_sha256": "c325652e444f8338730e18103aa632e41e221b3dd6a8f7f2de8bf401f7db6617", "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": 30.00096456700703, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.6396484375, "load_per_core": 0.051239013671875, "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:         +38.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.4\u00b0C  ", "Tccd1:        +43.9\u00b0C  ", "Tccd2:        +31.4\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:     +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)"]}, "machine_before": {"load_1m": 0.6669921875, "load_per_core": 0.020843505859375, "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:         +39.5\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +28.1\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:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 0, "rows": [{"decode_tps": [41.16224161849934, 41.107830335138004, 41.0301648643312], "native": {"agg": "mean", "decode_ms": 1557.18, "decode_ms_best": 1554.82, "decode_ms_worst": 1559.83, "decode_n": 64, "decode_tps": 41.1, "prefill_ms": 219.03, "prefill_ms_best": 218.71, "prefill_ms_worst": 219.6, "prefill_tps": 584.407, "repeats": 3, "repeats_requested": 3, "rss_mb": 906.07, "samples": {"decode_ms": [1554.823, 1556.881, 1559.828], "prefill_ms": [218.763, 218.712, 219.601], "total_ms": [1773.585, 1775.594, 1779.429], "ttft_ms": [218.763, 218.713, 219.602]}, "seq_len": 128, "threads": 4, "total_ms": 1776.2, "total_ms_best": 1773.59, "total_ms_worst": 1779.43, "total_tps": 108.096, "total_tps_best": 108.255, "total_tps_worst": 107.9, "ttft_ms": 219.03, "warmup": 64}, "prefill_tps": [585.108085005234, 585.2445224770474, 582.875305668007], "seq_len": 128}, {"decode_tps": [40.97347863553194, 40.935478011252144, 40.85642732248042], "native": {"agg": "mean", "decode_ms": 1563.96, "decode_ms_best": 1561.99, "decode_ms_worst": 1566.46, "decode_n": 64, "decode_tps": 40.922, "prefill_ms": 426.51, "prefill_ms_best": 425.76, "prefill_ms_worst": 427.92, "prefill_tps": 600.222, "repeats": 3, "repeats_requested": 3, "rss_mb": 916.82, "samples": {"decode_ms": [1561.986, 1563.436, 1566.461], "prefill_ms": [425.844, 425.758, 427.924], "total_ms": [1987.83, 1989.194, 1994.385], "ttft_ms": [425.845, 425.758, 427.924]}, "seq_len": 256, "threads": 4, "total_ms": 1990.47, "total_ms_best": 1987.83, "total_ms_worst": 1994.39, "total_tps": 160.766, "total_tps_best": 160.98, "total_tps_worst": 160.45, "ttft_ms": 426.51, "warmup": 64}, "prefill_tps": [601.1591099087929, 601.2805396492844, 598.237070133949], "seq_len": 256}, {"decode_tps": [40.66572330660621, 40.636055864417806, 40.59867825928092], "native": {"agg": "mean", "decode_ms": 1575.06, "decode_ms_best": 1573.81, "decode_ms_worst": 1576.41, "decode_n": 64, "decode_tps": 40.633, "prefill_ms": 791.83, "prefill_ms_best": 791.52, "prefill_ms_worst": 792.08, "prefill_tps": 646.599, "repeats": 3, "repeats_requested": 3, "rss_mb": 930.82, "samples": {"decode_ms": [1573.807, 1574.956, 1576.406], "prefill_ms": [792.083, 791.899, 791.523], "total_ms": [2365.889, 2366.855, 2367.929], "ttft_ms": [792.083, 791.9, 791.523]}, "seq_len": 512, "threads": 4, "total_ms": 2366.89, "total_ms_best": 2365.89, "total_ms_worst": 2367.93, "total_tps": 243.357, "total_tps_best": 243.46, "total_tps_worst": 243.251, "ttft_ms": 791.84, "warmup": 64}, "prefill_tps": [646.3969053748156, 646.5470975465306, 646.8542291253697], "seq_len": 512}, {"decode_tps": [40.359451363708025, 40.246408636879295, 40.21595970360838], "native": {"agg": "mean", "decode_ms": 1589.12, "decode_ms_best": 1585.75, "decode_ms_worst": 1591.41, "decode_n": 64, "decode_tps": 40.274, "prefill_ms": 1604.49, "prefill_ms_best": 1601.12, "prefill_ms_worst": 1607.32, "prefill_tps": 638.21, "repeats": 3, "repeats_requested": 3, "rss_mb": 932.32, "samples": {"decode_ms": [1585.75, 1590.204, 1591.408], "prefill_ms": [1601.122, 1605.023, 1607.317], "total_ms": [3186.871, 3195.227, 3198.725], "ttft_ms": [1601.122, 1605.023, 1607.317]}, "seq_len": 1024, "threads": 4, "total_ms": 3193.61, "total_ms_best": 3186.87, "total_ms_worst": 3198.72, "total_tps": 340.681, "total_tps_best": 341.401, "total_tps_worst": 340.136, "ttft_ms": 1604.49, "warmup": 64}, "prefill_tps": [639.5515145004565, 637.9970878921985, 637.086523691344], "seq_len": 1024}], "stderr": ["[bench] backend: vulkan", "[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": 47.16068403198733, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.513671875, "load_per_core": 0.01605224609375, "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:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.1\u00b0C  ", "Tccd1:        +37.8\u00b0C  ", "Tccd2:        +30.0\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:     +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:    +31.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +31.9\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": 0.4501953125, "load_per_core": 0.014068603515625, "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:         +39.8\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +29.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +29.9\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:    +31.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +31.9\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": [29.3935, 29.3777, 29.3865], "native": {"decode": {"avg_ns": 2177914674, "avg_ts": 29.385909, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2177349555, 2178524624, 2177869844], "samples_ts": [29.3935, 29.3777, 29.3865], "split_mode": "layer", "stddev_ns": 590662, "stddev_ts": 0.007944, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:30:10Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 242892510, "avg_ts": 526.985373, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [242395080, 242534016, 243748436], "samples_ts": [528.064, 527.761, 525.132], "split_mode": "layer", "stddev_ns": 744827, "stddev_ts": 1.612546, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:29:55Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [528.064, 527.761, 525.132], "seq_len": 128}, {"decode_tps": [29.5656, 29.5336, 29.5097], "native": {"decode": {"avg_ns": 2166828439, "avg_ts": 29.536274, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2164680978, 2167023992, 2168780348], "samples_ts": [29.5656, 29.5336, 29.5097], "split_mode": "layer", "stddev_ns": 2057196, "stddev_ts": 0.028038, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:30:17Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 469982296, "avg_ts": 544.70137, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [469947837, 469988901, 470010152], "samples_ts": [544.741, 544.694, 544.669], "split_mode": "layer", "stddev_ns": 44084, "stddev_ts": 0.036715, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:29:56Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [544.741, 544.694, 544.669], "seq_len": 256}, {"decode_tps": [29.3877, 29.4149, 29.3606], "native": {"decode": {"avg_ns": 2177778949, "avg_ts": 29.387756, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2177782501, 2175765806, 2179788540], "samples_ts": [29.3877, 29.4149, 29.3606], "split_mode": "layer", "stddev_ns": 2011369, "stddev_ts": 0.027142, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:30:24Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 920682309, "avg_ts": 556.109473, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [920472116, 921378384, 920196428], "samples_ts": [556.236, 555.689, 556.403], "split_mode": "layer", "stddev_ns": 619121, "stddev_ts": 0.373399, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:29:58Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [556.236, 555.689, 556.403], "seq_len": 512}, {"decode_tps": [29.2598, 29.2797, 29.2266], "native": {"decode": {"avg_ns": 2187635180, "avg_ts": 29.255351, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2187300985, 2185816352, 2189788204], "samples_ts": [29.2598, 29.2797, 29.2266], "split_mode": "layer", "stddev_ns": 2007449, "stddev_ts": 0.026832, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:30:32Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1855572526, "avg_ts": 551.851591, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [1853678224, 1855751104, 1857288251], "samples_ts": [552.415, 551.798, 551.341], "split_mode": "layer", "stddev_ns": 1812138, "stddev_ts": 0.538859, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:30:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [552.415, 551.798, 551.341], "seq_len": 1024}], "stderr": {"decode": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"], "prefill": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"]}}
{"cycle": 1, "elapsed_seconds": 47.55114791501546, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.29345703125, "load_per_core": 0.0091705322265625, "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:          22.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.4\u00b0C  ", "Tccd1:        +34.4\u00b0C  ", "Tccd2:        +30.5\u00b0C  ", "", "nvme-pci-0200", "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)", "", "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:    +33.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +36.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.1533203125, "load_per_core": 0.004791259765625, "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:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.9\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +27.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +31.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +31.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +34.9\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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [29.1581, 29.1015, 29.071], "native": {"decode": {"avg_ns": 2198543546, "avg_ts": 29.110225, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2194930384, 2199195966, 2201504288], "samples_ts": [29.1581, 29.1015, 29.071], "split_mode": "layer", "stddev_ns": 3335160, "stddev_ts": 0.044179, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:12Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 248518468, "avg_ts": 515.053679, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [247936367, 248809444, 248809594], "samples_ts": [516.261, 514.45, 514.45], "split_mode": "layer", "stddev_ns": 504360, "stddev_ts": 1.046001, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:44Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [516.261, 514.45, 514.45], "seq_len": 128}, {"decode_tps": [29.2928, 29.31, 29.2717], "native": {"decode": {"avg_ns": 2184933261, "avg_ts": 29.291521, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2184838483, 2183552053, 2186409247], "samples_ts": [29.2928, 29.31, 29.2717], "split_mode": "layer", "stddev_ns": 1430953, "stddev_ts": 0.019182, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:19Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 481079392, "avg_ts": 532.136963, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [480687863, 481036516, 481513798], "samples_ts": [532.57, 532.184, 531.657], "split_mode": "layer", "stddev_ns": 415213, "stddev_ts": 0.458579, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:45Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [532.57, 532.184, 531.657], "seq_len": 256}, {"decode_tps": [29.3279, 29.3569, 29.3908], "native": {"decode": {"avg_ns": 2179947611, "avg_ts": 29.358526, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2182225054, 2180064630, 2177553150], "samples_ts": [29.3279, 29.3569, 29.3908], "split_mode": "layer", "stddev_ns": 2338615, "stddev_ts": 0.031492, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:26Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 940211904, "avg_ts": 544.558294, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [939593612, 940064650, 940977452], "samples_ts": [544.916, 544.643, 544.115], "split_mode": "layer", "stddev_ns": 704908, "stddev_ts": 0.407409, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:47Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [544.916, 544.643, 544.115], "seq_len": 512}, {"decode_tps": [29.4737, 29.4579, 29.4122], "native": {"decode": {"avg_ns": 2173328740, "avg_ts": 29.447938, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2171424866, 2172591298, 2175970057], "samples_ts": [29.4737, 29.4579, 29.4122], "split_mode": "layer", "stddev_ns": 2361086, "stddev_ts": 0.031971, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:34Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1913390453, "avg_ts": 535.175835, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [1913192697, 1912163180, 1914815482], "samples_ts": [535.231, 535.519, 534.777], "split_mode": "layer", "stddev_ns": 1337163, "stddev_ts": 0.373948, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:32:50Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [535.231, 535.519, 534.777], "seq_len": 1024}], "stderr": {"decode": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"], "prefill": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"]}}
{"cycle": 1, "elapsed_seconds": 30.459274519991595, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.2734375, "load_per_core": 0.039794921875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.39 V  ", "vddnb:         1.19 V  ", "edge:         +38.0\u00b0C  ", "PPT:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.5\u00b0C  ", "Tccd1:        +45.4\u00b0C  ", "Tccd2:        +32.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +33.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +36.9\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:    +34.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.06298828125, "load_per_core": 0.0019683837890625, "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:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.1\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +28.1\u00b0C  ", "", "nvme-pci-0200", "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)", "", "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:    +34.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 1, "rows": [{"decode_tps": [40.62968392010427, 40.61875050376771, 40.58238255363374], "native": {"agg": "mean", "decode_ms": 1575.96, "decode_ms_best": 1575.2, "decode_ms_worst": 1577.04, "decode_n": 64, "decode_tps": 40.61, "prefill_ms": 222.89, "prefill_ms_best": 222.25, "prefill_ms_worst": 223.58, "prefill_tps": 574.285, "repeats": 3, "repeats_requested": 3, "rss_mb": 906.21, "samples": {"decode_ms": [1575.203, 1575.627, 1577.039], "prefill_ms": [223.582, 222.248, 222.827], "total_ms": [1798.785, 1797.875, 1799.866], "ttft_ms": [223.583, 222.249, 222.827]}, "seq_len": 128, "threads": 4, "total_ms": 1798.84, "total_ms_best": 1797.87, "total_ms_worst": 1799.87, "total_tps": 106.735, "total_tps_best": 106.793, "total_tps_worst": 106.675, "ttft_ms": 222.89, "warmup": 64}, "prefill_tps": [572.4968915207843, 575.933191749757, 574.4366706009595], "seq_len": 128}, {"decode_tps": [40.40636172911449, 40.40761178386979, 40.20353037251083], "native": {"agg": "mean", "decode_ms": 1586.56, "decode_ms_best": 1583.86, "decode_ms_worst": 1591.9, "decode_n": 64, "decode_tps": 40.339, "prefill_ms": 432.98, "prefill_ms_best": 432.06, "prefill_ms_worst": 433.71, "prefill_tps": 591.248, "repeats": 3, "repeats_requested": 3, "rss_mb": 916.96, "samples": {"decode_ms": [1583.909, 1583.86, 1591.9], "prefill_ms": [433.714, 432.059, 433.176], "total_ms": [2017.623, 2015.919, 2025.075], "ttft_ms": [433.714, 432.059, 433.176]}, "seq_len": 256, "threads": 4, "total_ms": 2019.54, "total_ms_best": 2015.92, "total_ms_worst": 2025.08, "total_tps": 158.452, "total_tps_best": 158.737, "total_tps_worst": 158.019, "ttft_ms": 432.98, "warmup": 64}, "prefill_tps": [590.2507182152294, 592.5116708597668, 590.9838033501394], "seq_len": 256}, {"decode_tps": [40.03818642029836, 39.95846816714877, 39.959366319373984], "native": {"agg": "mean", "decode_ms": 1600.59, "decode_ms_best": 1598.47, "decode_ms_worst": 1601.66, "decode_n": 64, "decode_tps": 39.985, "prefill_ms": 806.5, "prefill_ms_best": 805.41, "prefill_ms_worst": 808.09, "prefill_tps": 634.843, "repeats": 3, "repeats_requested": 3, "rss_mb": 931.21, "samples": {"decode_ms": [1598.474, 1601.663, 1601.627], "prefill_ms": [805.411, 805.998, 808.086], "total_ms": [2403.885, 2407.662, 2409.713], "ttft_ms": [805.411, 805.999, 808.087]}, "seq_len": 512, "threads": 4, "total_ms": 2407.09, "total_ms_best": 2403.89, "total_ms_worst": 2409.71, "total_tps": 239.293, "total_tps_best": 239.612, "total_tps_worst": 239.033, "ttft_ms": 806.5, "warmup": 64}, "prefill_tps": [635.7002822161605, 635.237308281162, 633.5959291461552], "seq_len": 512}, {"decode_tps": [39.761233791084535, 39.66301312473894, 39.602539017782775], "native": {"agg": "mean", "decode_ms": 1613.09, "decode_ms_best": 1609.61, "decode_ms_worst": 1616.06, "decode_n": 64, "decode_tps": 39.675, "prefill_ms": 1633.47, "prefill_ms_best": 1630.7, "prefill_ms_worst": 1636.82, "prefill_tps": 626.887, "repeats": 3, "repeats_requested": 3, "rss_mb": 932.71, "samples": {"decode_ms": [1609.608, 1613.594, 1616.058], "prefill_ms": [1630.705, 1632.873, 1636.824], "total_ms": [3240.313, 3246.467, 3252.882], "ttft_ms": [1630.705, 1632.874, 1636.824]}, "seq_len": 1024, "threads": 4, "total_ms": 3246.55, "total_ms_best": 3240.31, "total_ms_worst": 3252.88, "total_tps": 335.125, "total_tps_best": 335.77, "total_tps_worst": 334.473, "ttft_ms": 1633.47, "warmup": 64}, "prefill_tps": [627.949261209109, 627.1155197005523, 625.601775145037], "seq_len": 1024}], "stderr": ["[bench] backend: vulkan", "[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": 30.40319641199312, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.353515625, "load_per_core": 0.04229736328125, "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:         +39.0\u00b0C  ", "PPT:          18.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.6\u00b0C  ", "Tccd1:        +40.6\u00b0C  ", "Tccd2:        +31.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +33.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +37.9\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:    +35.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.2802734375, "load_per_core": 0.008758544921875, "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:         +37.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.4\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +28.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +33.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +37.9\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:    +35.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +39.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 0, "rows": [{"decode_tps": [40.64986429932019, 40.63494643488662, 40.61539945562672], "native": {"agg": "mean", "decode_ms": 1575.06, "decode_ms_best": 1574.42, "decode_ms_worst": 1575.76, "decode_n": 64, "decode_tps": 40.633, "prefill_ms": 221.84, "prefill_ms_best": 220.96, "prefill_ms_worst": 222.38, "prefill_tps": 576.99, "repeats": 3, "repeats_requested": 3, "rss_mb": 906.55, "samples": {"decode_ms": [1574.421, 1574.999, 1575.757], "prefill_ms": [220.964, 222.178, 222.381], "total_ms": [1795.385, 1797.177, 1798.138], "ttft_ms": [220.964, 222.178, 222.381]}, "seq_len": 128, "threads": 4, "total_ms": 1796.9, "total_ms_best": 1795.38, "total_ms_worst": 1798.14, "total_tps": 106.851, "total_tps_best": 106.941, "total_tps_worst": 106.777, "ttft_ms": 221.84, "warmup": 64}, "prefill_tps": [579.2798826958237, 576.1146468147161, 575.5887418439524], "seq_len": 128}, {"decode_tps": [40.41513925857164, 40.421086670388775, 40.39633857686224], "native": {"agg": "mean", "decode_ms": 1583.73, "decode_ms_best": 1583.33, "decode_ms_worst": 1584.3, "decode_n": 64, "decode_tps": 40.411, "prefill_ms": 432.87, "prefill_ms_best": 431.07, "prefill_ms_worst": 433.85, "prefill_tps": 591.402, "repeats": 3, "repeats_requested": 3, "rss_mb": 917.05, "samples": {"decode_ms": [1583.565, 1583.332, 1584.302], "prefill_ms": [431.075, 433.848, 433.686], "total_ms": [2014.64, 2017.18, 2017.989], "ttft_ms": [431.075, 433.849, 433.686]}, "seq_len": 256, "threads": 4, "total_ms": 2016.6, "total_ms_best": 2014.64, "total_ms_worst": 2017.99, "total_tps": 158.683, "total_tps_best": 158.837, "total_tps_worst": 158.574, "ttft_ms": 432.87, "warmup": 64}, "prefill_tps": [593.8641767673839, 590.0684110564068, 590.2888264781432], "seq_len": 256}, {"decode_tps": [40.03911320871577, 39.97224427288302, 39.96575434424627], "native": {"agg": "mean", "decode_ms": 1600.31, "decode_ms_best": 1598.44, "decode_ms_worst": 1601.37, "decode_n": 64, "decode_tps": 39.992, "prefill_ms": 805.09, "prefill_ms_best": 803.82, "prefill_ms_worst": 806.03, "prefill_tps": 635.95, "repeats": 3, "repeats_requested": 3, "rss_mb": 931.3, "samples": {"decode_ms": [1598.437, 1601.111, 1601.371], "prefill_ms": [803.816, 805.433, 806.034], "total_ms": [2402.253, 2406.544, 2407.405], "ttft_ms": [803.817, 805.434, 806.034]}, "seq_len": 512, "threads": 4, "total_ms": 2405.4, "total_ms_best": 2402.25, "total_ms_worst": 2407.4, "total_tps": 239.461, "total_tps_best": 239.775, "total_tps_worst": 239.262, "ttft_ms": 805.1, "warmup": 64}, "prefill_tps": [636.9616927256984, 635.6829183805481, 635.2089365957268], "seq_len": 512}, {"decode_tps": [39.75182439134677, 39.72133004391069, 39.56187693907285], "native": {"agg": "mean", "decode_ms": 1612.98, "decode_ms_best": 1609.99, "decode_ms_worst": 1617.72, "decode_n": 64, "decode_tps": 39.678, "prefill_ms": 1630.83, "prefill_ms_best": 1629.43, "prefill_ms_worst": 1632.05, "prefill_tps": 627.9, "repeats": 3, "repeats_requested": 3, "rss_mb": 932.8, "samples": {"decode_ms": [1609.989, 1611.225, 1617.719], "prefill_ms": [1629.427, 1632.049, 1631.02], "total_ms": [3239.416, 3243.274, 3248.74], "ttft_ms": [1629.427, 1632.05, 1631.021]}, "seq_len": 1024, "threads": 4, "total_ms": 3243.81, "total_ms_best": 3239.42, "total_ms_worst": 3248.74, "total_tps": 335.408, "total_tps_best": 335.863, "total_tps_worst": 334.899, "ttft_ms": 1630.83, "warmup": 64}, "prefill_tps": [628.4417773855472, 627.4321420496566, 627.8279849419382], "seq_len": 1024}], "stderr": ["[bench] backend: vulkan", "[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": 47.61500790700666, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.40087890625, "load_per_core": 0.0125274658203125, "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:         +37.0\u00b0C  ", "PPT:          23.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.6\u00b0C  ", "Tccd1:        +36.1\u00b0C  ", "Tccd2:        +30.5\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +34.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\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:    +36.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +36.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +40.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.4853515625, "load_per_core": 0.015167236328125, "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:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.1\u00b0C  ", "Tccd1:        +26.8\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +34.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +36.9\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:    +36.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +36.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +40.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [28.7861, 28.7762, 28.7772], "native": {"decode": {"avg_ns": 2223778129, "avg_ts": 28.77985, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2223295346, 2224058880, 2223980161], "samples_ts": [28.7861, 28.7762, 28.7772], "split_mode": "layer", "stddev_ns": 419950, "stddev_ts": 0.005436, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:46Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 244461088, "avg_ts": 523.613395, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [243791474, 243438619, 246153173], "samples_ts": [525.039, 525.8, 520.001], "split_mode": "layer", "stddev_ns": 1476136, "stddev_ts": 3.151119, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:31Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [525.039, 525.8, 520.001], "seq_len": 128}, {"decode_tps": [29.0339, 29.0253, 29.0001], "native": {"decode": {"avg_ns": 2205392979, "avg_ts": 29.019778, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2204323331, 2204969668, 2206885940], "samples_ts": [29.0339, 29.0253, 29.0001], "split_mode": "layer", "stddev_ns": 1334371, "stddev_ts": 0.017532, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:53Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 473851030, "avg_ts": 540.254676, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [474489049, 473472389, 473591652], "samples_ts": [539.528, 540.686, 540.55], "split_mode": "layer", "stddev_ns": 555749, "stddev_ts": 0.633223, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:32Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [539.528, 540.686, 540.55], "seq_len": 256}, {"decode_tps": [29.1449, 29.1707, 29.1099], "native": {"decode": {"avg_ns": 2196157319, "avg_ts": 29.141832, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2195924007, 2193983659, 2198564291], "samples_ts": [29.1449, 29.1707, 29.1099], "split_mode": "layer", "stddev_ns": 2299211, "stddev_ts": 0.030505, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:39:01Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 926405595, "avg_ts": 552.673855, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [926503001, 926968962, 925744823], "samples_ts": [552.616, 552.338, 553.068], "split_mode": "layer", "stddev_ns": 618604, "stddev_ts": 0.368656, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:34Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [552.616, 552.338, 553.068], "seq_len": 512}, {"decode_tps": [29.2535, 29.2482, 29.2209], "native": {"decode": {"avg_ns": 2188717684, "avg_ts": 29.240873, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2187772576, 2188169625, 2190210851], "samples_ts": [29.2535, 29.2482, 29.2209], "split_mode": "layer", "stddev_ns": 1308270, "stddev_ts": 0.017473, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:39:08Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1886873090, "avg_ts": 542.698099, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [1883879513, 1885917850, 1890821908], "samples_ts": [543.559, 542.972, 541.563], "split_mode": "layer", "stddev_ns": 3568677, "stddev_ts": 1.025617, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:38:37Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [543.559, 542.972, 541.563], "seq_len": 1024}], "stderr": {"decode": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"], "prefill": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"]}}
{"cycle": 3, "elapsed_seconds": 47.665460948017426, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.1943359375, "load_per_core": 0.006072998046875, "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:         +38.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.8\u00b0C  ", "Tccd1:        +36.6\u00b0C  ", "Tccd2:        +30.5\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +35.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\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:    +37.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +37.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +40.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.14990234375, "load_per_core": 0.0046844482421875, "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:          17.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +40.2\u00b0C  ", "Tccd1:        +29.6\u00b0C  ", "Tccd2:        +30.0\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +34.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +34.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\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:    +37.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +37.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +40.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [29.0433, 29.0289, 29.0184], "native": {"decode": {"avg_ns": 2204600770, "avg_ts": 29.030203, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2203603234, 2204701391, 2205497687], "samples_ts": [29.0433, 29.0289, 29.0184], "split_mode": "layer", "stddev_ns": 953541, "stddev_ts": 0.012527, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:40:49Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 249310520, "avg_ts": 513.424429, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [247913138, 249745105, 250273317], "samples_ts": [516.31, 512.523, 511.441], "split_mode": "layer", "stddev_ns": 1238652, "stddev_ts": 2.556726, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:20Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [516.31, 512.523, 511.441], "seq_len": 128}, {"decode_tps": [29.2514, 29.1472, 29.0754], "native": {"decode": {"avg_ns": 2194952892, "avg_ts": 29.15798, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2187928292, 2195754273, 2201176113], "samples_ts": [29.2514, 29.1472, 29.0754], "split_mode": "layer", "stddev_ns": 6660498, "stddev_ts": 0.088523, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:40:56Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 482624355, "avg_ts": 530.43443, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [483533600, 481763536, 482575931], "samples_ts": [529.436, 531.381, 530.486], "split_mode": "layer", "stddev_ns": 886569, "stddev_ts": 0.973653, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:21Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [529.436, 531.381, 530.486], "seq_len": 256}, {"decode_tps": [29.3303, 29.3469, 29.3309], "native": {"decode": {"avg_ns": 2181618262, "avg_ts": 29.336023, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2182047544, 2180805845, 2182001397], "samples_ts": [29.3303, 29.3469, 29.3309], "split_mode": "layer", "stddev_ns": 703952, "stddev_ts": 0.009468, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:03Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 941537156, "avg_ts": 543.791853, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [941307280, 942407347, 940896843], "samples_ts": [543.924, 543.289, 544.162], "split_mode": "layer", "stddev_ns": 782253, "stddev_ts": 0.45095, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:23Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [543.924, 543.289, 544.162], "seq_len": 512}, {"decode_tps": [29.4333, 29.3682, 29.3576], "native": {"decode": {"avg_ns": 2177882065, "avg_ts": 29.386386, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [2174410241, 2179224385, 2180011571], "samples_ts": [29.4333, 29.3682, 29.3576], "split_mode": "layer", "stddev_ns": 3033058, "stddev_ts": 0.040946, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:10Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 1920455021, "avg_ts": 533.210392, "backends": "Vulkan", "build_commit": "adfffbe41", "build_number": 10743, "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": "NVIDIA GeForce RTX 2080 Ti", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/geist-models/Ternary-Bonsai-2-27B-PQ2_0.gguf", "model_n_params": 26895998464, "model_size": 7195047936, "model_type": "qwen35 27B PQ2_0 - 2.13 bpw (group 128)", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "samples_ns": [1915150080, 1919306617, 1926908368], "samples_ts": [534.684, 533.526, 531.421], "split_mode": "layer", "stddev_ns": 5962994, "stddev_ts": 1.654101, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-12T01:41:27Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [534.684, 533.526, 531.421], "seq_len": 1024}], "stderr": {"decode": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"], "prefill": ["ggml_vulkan: Found 1 Vulkan devices:", "ggml_vulkan: 0 = NVIDIA GeForce RTX 2080 Ti (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | fp4: 0 | warp size: 32 | shared memory: 49152 | int dot: 0 | matrix cores: KHR_coopmat"]}}
{"cycle": 3, "elapsed_seconds": 30.467254232004052, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.29736328125, "load_per_core": 0.0405426025390625, "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:         +39.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.9\u00b0C  ", "Tccd1:        +40.6\u00b0C  ", "Tccd2:        +31.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +35.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\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:    +37.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +37.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +41.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.18798828125, "load_per_core": 0.0058746337890625, "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:         +40.2\u00b0C  ", "Tccd1:        +28.2\u00b0C  ", "Tccd2:        +29.1\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +35.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +38.9\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:    +38.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +38.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +41.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 1, "rows": [{"decode_tps": [40.52749062010227, 40.48483115611394, 40.49474454143498], "native": {"agg": "mean", "decode_ms": 1580.16, "decode_ms_best": 1579.17, "decode_ms_worst": 1580.84, "decode_n": 64, "decode_tps": 40.502, "prefill_ms": 221.84, "prefill_ms_best": 220.86, "prefill_ms_worst": 222.35, "prefill_tps": 576.982, "repeats": 3, "repeats_requested": 3, "rss_mb": 907.11, "samples": {"decode_ms": [1579.175, 1580.839, 1580.452], "prefill_ms": [220.861, 222.322, 222.35], "total_ms": [1800.035, 1803.16, 1802.802], "ttft_ms": [220.861, 222.322, 222.35]}, "seq_len": 128, "threads": 4, "total_ms": 1802.0, "total_ms_best": 1800.04, "total_ms_worst": 1803.16, "total_tps": 106.548, "total_tps_best": 106.665, "total_tps_worst": 106.48, "ttft_ms": 221.84, "warmup": 64}, "prefill_tps": [579.5500337316231, 575.7414920700605, 575.66899033056], "seq_len": 128}, {"decode_tps": [40.38940434463774, 40.22958521411883, 40.17796326604152], "native": {"agg": "mean", "decode_ms": 1589.45, "decode_ms_best": 1584.57, "decode_ms_worst": 1592.91, "decode_n": 64, "decode_tps": 40.265, "prefill_ms": 433.82, "prefill_ms_best": 432.62, "prefill_ms_worst": 434.97, "prefill_tps": 590.103, "repeats": 3, "repeats_requested": 3, "rss_mb": 917.86, "samples": {"decode_ms": [1584.574, 1590.869, 1592.913], "prefill_ms": [432.621, 433.873, 434.973], "total_ms": [2017.195, 2024.743, 2027.886], "ttft_ms": [432.621, 433.874, 434.973]}, "seq_len": 256, "threads": 4, "total_ms": 2023.27, "total_ms_best": 2017.19, "total_ms_worst": 2027.89, "total_tps": 158.159, "total_tps_best": 158.636, "total_tps_worst": 157.8, "ttft_ms": 433.82, "warmup": 64}, "prefill_tps": [591.7419635200325, 590.0344109912347, 588.5422773367542], "seq_len": 256}, {"decode_tps": [39.987503905029676, 39.94305618053263, 39.94203412297903], "native": {"agg": "mean", "decode_ms": 1601.7, "decode_ms_best": 1600.5, "decode_ms_worst": 1602.32, "decode_n": 64, "decode_tps": 39.958, "prefill_ms": 807.92, "prefill_ms_best": 805.77, "prefill_ms_worst": 809.34, "prefill_tps": 633.727, "repeats": 3, "repeats_requested": 3, "rss_mb": 931.86, "samples": {"decode_ms": [1600.5, 1602.281, 1602.322], "prefill_ms": [809.336, 805.771, 808.65], "total_ms": [2409.836, 2408.052, 2410.972], "ttft_ms": [809.337, 805.771, 808.651]}, "seq_len": 512, "threads": 4, "total_ms": 2409.62, "total_ms_best": 2408.05, "total_ms_worst": 2410.97, "total_tps": 239.042, "total_tps_best": 239.197, "total_tps_worst": 238.908, "ttft_ms": 807.92, "warmup": 64}, "prefill_tps": [632.6173554617612, 635.416265911779, 633.1540221356582], "seq_len": 512}, {"decode_tps": [39.71097853435465, 39.58443762575017, 39.53926867180183], "native": {"agg": "mean", "decode_ms": 1615.7, "decode_ms_best": 1611.64, "decode_ms_worst": 1618.64, "decode_n": 64, "decode_tps": 39.611, "prefill_ms": 1634.74, "prefill_ms_best": 1632.06, "prefill_ms_worst": 1637.59, "prefill_tps": 626.4, "repeats": 3, "repeats_requested": 3, "rss_mb": 933.36, "samples": {"decode_ms": [1611.645, 1616.797, 1618.644], "prefill_ms": [1632.06, 1634.564, 1637.588], "total_ms": [3243.705, 3251.361, 3256.232], "ttft_ms": [1632.061, 1634.565, 1637.589]}, "seq_len": 1024, "threads": 4, "total_ms": 3250.43, "total_ms_best": 3243.7, "total_ms_worst": 3256.23, "total_tps": 334.725, "total_tps_best": 335.419, "total_tps_worst": 334.128, "ttft_ms": 1634.74, "warmup": 64}, "prefill_tps": [627.427913189466, 626.4667519901332, 625.3099070095775], "seq_len": 1024}], "stderr": ["[bench] backend: vulkan", "[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": 0.3088566090090127, "mad_tps": 0.12266596736620272, "median_tps": 39.716154289132675, "samples": 12, "vs_llama_pct": 35.46378939389323}, "prefill": {"mad_pct": 0.15733992922625994, "mad_tps": 0.9875126977070181, "median_tps": 627.6300634957975, "samples": 12, "vs_llama_pct": 16.54267056654879}}, "128": {"decode": {"mad_pct": 0.170540209320324, "mad_tps": 0.06928062506797872, "median_tps": 40.624217211935985, "samples": 12, "vs_llama_pct": 39.807989468808834}, "prefill": {"mad_pct": 0.42040025880342163, "mad_tps": 2.4216060474321353, "median_tps": 576.0239192822366, "samples": 12, "vs_llama_pct": 11.168156910857196}}, "256": {"decode": {"mad_pct": 0.2412750490438156, "mad_tps": 0.09749197711385449, "median_tps": 40.40698675649214, "samples": 12, "vs_llama_pct": 38.08901700864151}, "prefill": {"mad_pct": 0.22177117435018892, "mad_tps": 1.3114724112651288, "median_tps": 591.3628834350859, "samples": 12, "vs_llama_pct": 10.318811048073195}}, "512": {"decode": {"mad_pct": 0.09336932156780281, "mad_tps": 0.03732893720052033, "median_tps": 39.97987408895635, "samples": 12, "vs_llama_pct": 36.26916513214997}, "prefill": {"mad_pct": 0.26479157733210923, "mad_tps": 1.6828817897716135, "median_tps": 635.5495921461636, "samples": 12, "vs_llama_pct": 15.843659197626735}}}, "llama.cpp": {"1024": {"decode": {"mad_pct": 0.2771273575011031, "mad_tps": 0.08124999999999716, "median_tps": 29.318649999999998, "samples": 12}, "prefill": {"mad_pct": 0.8769991514109443, "mad_tps": 4.723000000000013, "median_tps": 538.5409999999999, "samples": 12}}, "128": {"decode": {"mad_pct": 0.640117836745861, "mad_tps": 0.18599999999999994, "median_tps": 29.05715, "samples": 12}, "prefill": {"mad_pct": 1.1914377054764376, "mad_tps": 6.173499999999962, "median_tps": 518.1555, "samples": 12}}, "256": {"decode": {"mad_pct": 0.7070712248667653, "mad_tps": 0.20690000000000097, "median_tps": 29.26155, "samples": 12}, "prefill": {"mad_pct": 0.8679243875093567, "mad_tps": 4.652500000000032, "median_tps": 536.049, "samples": 12}}, "512": {"decode": {"mad_pct": 0.1201476537975104, "mad_tps": 0.03524999999999778, "median_tps": 29.338900000000002, "samples": 12}, "prefill": {"mad_pct": 0.8181332672289062, "mad_tps": 4.488499999999931, "median_tps": 548.627, "samples": 12}}}}}