{"created_utc": "2026-10-09T18:48:57.888210+00:00", "engines": [{"binary": "/home/germar/workplace/geistlib-wt/hd96/bin/linux/release/tests/bench_perf_sweep", "binary_sha256": "b1a3283699b67735d3ea8fc0d943c3765f26da65624707a46012948ae14d77cd", "commit": "db7f140d5728a7107dabaecfaa304d2cc302e574", "kind": "geist", "label": "geist", "linked_libraries": ["linux-vdso.so.1 (0x000073dc741eb000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x000073dc73cda000)", "\tlibmvec.so.1 => /usr/lib/x86_64-linux-gnu/libmvec.so.1 (0x000073dc740cb000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x000073dc73c7b000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x000073dc73a00000)", "\t/lib64/ld-linux-x86-64.so.2 (0x000073dc741ed000)"]}, {"binary": "/home/germar/llama.cpp-v0.6.0/build-vulkan/bin/llama-bench", "binary_sha256": "3b4b3f9c50de03a4cafb0b1f498a3a9b2d8b306db54e2a555cb4129857fc0f8f", "commit": "d81235049384534c167caea52b85a694f6103d14", "kind": "llama", "label": "llama.cpp", "linked_libraries": ["linux-vdso.so.1 (0x0000733b27912000)", "\tlibllama-bench-impl.so => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama-bench-impl.so (0x0000733b2789f000)", "\tlibstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x0000733b27400000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x0000733b27854000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x0000733b27000000)", "\tlibllama-common.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama-common.so.0 (0x0000733b26a00000)", "\tlibllama.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama.so.0 (0x0000733b26400000)", "\tlibggml.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml.so.0 (0x0000733b27844000)", "\tlibggml-base.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-base.so.0 (0x0000733b27769000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x0000733b272da000)", "\t/lib64/ld-linux-x86-64.so.2 (0x0000733b27914000)", "\tlibssl.so.3 => /usr/lib/x86_64-linux-gnu/libssl.so.3 (0x0000733b268ef000)", "\tlibcrypto.so.3 => /usr/lib/x86_64-linux-gnu/libcrypto.so.3 (0x0000733b25c00000)", "\tlibggml-cpu.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-cpu.so.0 (0x0000733b25800000)", "\tlibggml-vulkan.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-vulkan.so.0 (0x0000733b23600000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x0000733b27708000)", "\tlibz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1 (0x0000733b276ea000)", "\tlibzstd.so.1 => /usr/lib/x86_64-linux-gnu/libzstd.so.1 (0x0000733b26f3d000)", "\tlibvulkan.so.1 => /usr/lib/x86_64-linux-gnu/libvulkan.so.1 (0x0000733b2724a000)"]}], "host_profile": "nvidia_2080ti_vulkan", "kind": "metadata", "model": {"file": "bitnet_b1_58-large-TQ2_0.gguf", "key": "bitnet-large-tq2", "path": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "sha256": "281aafb18a9f4a3124c10a1d8683e2296f0cfe8a2944da0a5667d17488a951bb"}, "protocol": {"host_gate": {"cooldown_seconds": 60, "max_load_per_core": 0.2, "settle_seconds": 30, "temp_gate": {"max_millic": 62000, "note": "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": "db7f140d5728a7107dabaecfaa304d2cc302e574", "path": "tools/bench_cross_engine.py", "sha256": "1b2af101ff1c7bc98722fb96a3f43dff448df3808c9012cc7d69475f04c067e5"}, "schema": "geist.benchmark.cross-engine-run.v1", "system": {"compiler": "cc (Ubuntu 15.2.0-16ubuntu1) 15.2.0", "cpu": "AMD Ryzen 9 9950X 16-Core Processor", "logical_cores": 32, "memory_bytes": "63813259264", "os": "Linux 7.0.0-34-generic", "performance_cores": "16", "product": "Ubuntu 26.04.1 LTS"}}
{"cycle": 0, "elapsed_seconds": 3.1658432588446885, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 3.24365234375, "load_per_core": 0.1013641357421875, "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:         +40.0\u00b0C  ", "PPT:          21.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +46.1\u00b0C  ", "Tccd1:        +43.2\u00b0C  ", "Tccd2:        +33.2\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:        +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:    +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)"]}, "machine_before": {"load_1m": 3.24365234375, "load_per_core": 0.1013641357421875, "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:         +39.0\u00b0C  ", "PPT:          18.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.2\u00b0C  ", "Tccd1:        +42.8\u00b0C  ", "Tccd2:        +32.9\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:     +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:     +37.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 0, "rows": [{"decode_tps": [266.11005313885124, 348.0227954931048, 349.1810612921882], "native": {"agg": "mean", "decode_ms": 202.56, "decode_ms_best": 183.29, "decode_ms_worst": 240.5, "decode_n": 64, "decode_tps": 315.954, "prefill_ms": 17.21, "prefill_ms_best": 14.89, "prefill_ms_worst": 21.79, "prefill_tps": 7436.303, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.69, "samples": {"decode_ms": [240.502, 183.896, 183.286], "prefill_ms": [21.788, 14.959, 14.891], "total_ms": [262.29, 198.856, 198.177], "ttft_ms": [21.788, 14.959, 14.891]}, "seq_len": 128, "threads": 4, "total_ms": 219.77, "total_ms_best": 198.18, "total_ms_worst": 262.29, "total_tps": 873.624, "total_tps_best": 968.83, "total_tps_worst": 732.015, "ttft_ms": 17.21, "warmup": 64}, "prefill_tps": [5874.793464292271, 8556.72170599639, 8595.796118460816], "seq_len": 128}, {"decode_tps": [348.10797874365653, 347.8468821505633, 349.2153567453129], "native": {"agg": "mean", "decode_ms": 183.7, "decode_ms_best": 183.27, "decode_ms_worst": 183.99, "decode_n": 64, "decode_tps": 348.389, "prefill_ms": 23.24, "prefill_ms_best": 23.23, "prefill_ms_worst": 23.25, "prefill_tps": 11015.178, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.69, "samples": {"decode_ms": [183.851, 183.989, 183.268], "prefill_ms": [23.229, 23.244, 23.249], "total_ms": [207.08, 207.233, 206.517], "ttft_ms": [23.229, 23.244, 23.25]}, "seq_len": 256, "threads": 4, "total_ms": 206.94, "total_ms_best": 206.52, "total_ms_worst": 207.23, "total_tps": 1546.316, "total_tps_best": 1549.513, "total_tps_worst": 1544.152, "ttft_ms": 23.24, "warmup": 64}, "prefill_tps": [11020.706875026906, 11013.594906212356, 11011.226289302766], "seq_len": 256}, {"decode_tps": [342.46941855114034, 340.7590407633002, 341.37338781083645], "native": {"agg": "mean", "decode_ms": 187.39, "decode_ms_best": 186.88, "decode_ms_worst": 187.82, "decode_n": 64, "decode_tps": 341.533, "prefill_ms": 38.64, "prefill_ms_best": 38.62, "prefill_ms_worst": 38.66, "prefill_tps": 13250.388, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.69, "samples": {"decode_ms": [186.878, 187.816, 187.478], "prefill_ms": [38.639, 38.621, 38.662], "total_ms": [225.517, 226.437, 226.139], "ttft_ms": [38.639, 38.621, 38.662]}, "seq_len": 512, "threads": 4, "total_ms": 226.03, "total_ms_best": 225.52, "total_ms_worst": 226.44, "total_tps": 2548.324, "total_tps_best": 2554.132, "total_tps_worst": 2543.758, "ttft_ms": 38.64, "warmup": 64}, "prefill_tps": [13250.860529516809, 13257.036327386653, 13242.977600744918], "seq_len": 512}, {"decode_tps": [322.56926418926747, 324.1983688769566, 321.0433910208177], "native": {"agg": "mean", "decode_ms": 198.39, "decode_ms_best": 197.41, "decode_ms_worst": 199.35, "decode_n": 64, "decode_tps": 322.599, "prefill_ms": 86.26, "prefill_ms_best": 85.88, "prefill_ms_worst": 86.93, "prefill_tps": 11871.023, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.69, "samples": {"decode_ms": [198.407, 197.41, 199.35], "prefill_ms": [85.877, 85.97, 86.935], "total_ms": [284.284, 283.38, 286.285], "ttft_ms": [85.878, 85.97, 86.935]}, "seq_len": 1024, "threads": 4, "total_ms": 284.65, "total_ms_best": 283.38, "total_ms_worst": 286.28, "total_tps": 3822.245, "total_tps_best": 3839.372, "total_tps_worst": 3800.413, "ttft_ms": 86.26, "warmup": 64}, "prefill_tps": [11924.030881376853, 11911.131790159357, 11778.91528153218], "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": 3.683761515887454, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 3.81298828125, "load_per_core": 0.1191558837890625, "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:          24.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.8\u00b0C  ", "Tccd1:        +47.2\u00b0C  ", "Tccd2:        +34.6\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:        +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:    +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": 3.79638671875, "load_per_core": 0.1186370849609375, "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:         +40.0\u00b0C  ", "PPT:          19.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +44.9\u00b0C  ", "Tccd1:        +41.9\u00b0C  ", "Tccd2:        +33.8\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:    +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [379.241, 381.223, 382.282], "native": {"decode": {"avg_ns": 168018201, "avg_ts": 380.915288, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [168758123, 167880655, 167415825], "samples_ts": [379.241, 381.223, 382.282], "split_mode": "layer", "stddev_ns": 681637, "stddev_ts": 1.543537, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:02Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 18448327, "avg_ts": 6938.327266, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [18402589, 18494953, 18447441], "samples_ts": [6955.54, 6920.81, 6938.63], "split_mode": "layer", "stddev_ns": 46586, "stddev_ts": 17.370031, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [6955.54, 6920.81, 6938.63], "seq_len": 128}, {"decode_tps": [375.03, 382.444, 375.834], "native": {"decode": {"avg_ns": 169428478, "avg_ts": 377.769543, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [170652935, 167344772, 170287729], "samples_ts": [375.03, 382.444, 375.834], "split_mode": "layer", "stddev_ns": 1813851, "stddev_ts": 4.068104, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:03Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 26972776, "avg_ts": 9492.433175, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [27433256, 26776718, 26708355], "samples_ts": [9331.74, 9560.54, 9585.02], "split_mode": "layer", "stddev_ns": 400283, "stddev_ts": 139.701617, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9331.74, 9560.54, 9585.02], "seq_len": 256}, {"decode_tps": [357.466, 360.552, 365.264], "native": {"decode": {"avg_ns": 177252980, "avg_ts": 361.094273, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [179037774, 177505638, 175215530], "samples_ts": [357.466, 360.552, 365.264], "split_mode": "layer", "stddev_ns": 1923699, "stddev_ts": 3.927169, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:03Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 37025687, "avg_ts": 13848.312844, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [39046354, 36012843, 36017864], "samples_ts": [13112.6, 14217.2, 14215.2], "split_mode": "layer", "stddev_ns": 1749950, "stddev_ts": 637.129516, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13112.6, 14217.2, 14215.2], "seq_len": 512}, {"decode_tps": [337.564, 345.88, 343.392], "native": {"decode": {"avg_ns": 187001575, "avg_ts": 342.278751, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [189593416, 185035121, 186376188], "samples_ts": [337.564, 345.88, 343.392], "split_mode": "layer", "stddev_ns": 2342614, "stddev_ts": 4.268123, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:04Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77663405, "avg_ts": 13185.650763, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [78372869, 77305253, 77312093], "samples_ts": [13065.7, 13246.2, 13245], "split_mode": "layer", "stddev_ns": 614423, "stddev_ts": 103.84222, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:52:01Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13065.7, 13246.2, 13245], "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": 3.618295517982915, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 1.6201171875, "load_per_core": 0.050628662109375, "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:         +39.0\u00b0C  ", "PPT:          21.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.4\u00b0C  ", "Tccd1:        +40.1\u00b0C  ", "Tccd2:        +32.4\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:    +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": 1.67431640625, "load_per_core": 0.0523223876953125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +39.0\u00b0C  ", "PPT:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.9\u00b0C  ", "Tccd1:        +38.8\u00b0C  ", "Tccd2:        +31.4\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:        +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:    +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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [289.715, 386.528, 387.211], "native": {"decode": {"avg_ns": 183922577, "avg_ts": 354.484748, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [220906529, 165576498, 165284704], "samples_ts": [289.715, 386.528, 387.211], "split_mode": "layer", "stddev_ns": 32029374, "stddev_ts": 56.09303, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:35Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 13210829, "avg_ts": 9689.280969, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [13301560, 13195036, 13135893], "samples_ts": [9622.93, 9700.62, 9744.29], "split_mode": "layer", "stddev_ns": 84112, "stddev_ts": 61.470264, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9622.93, 9700.62, 9744.29], "seq_len": 128}, {"decode_tps": [375.03, 385.06, 381.516], "native": {"decode": {"avg_ns": 168204353, "avg_ts": 380.535053, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [170653115, 166207965, 167751981], "samples_ts": [375.03, 385.06, 381.516], "split_mode": "layer", "stddev_ns": 2256913, "stddev_ts": 5.086382, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:35Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 20165791, "avg_ts": 12721.180378, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [19511846, 21486401, 19499126], "samples_ts": [13120.2, 11914.5, 13128.8], "split_mode": "layer", "stddev_ns": 1143699, "stddev_ts": 698.607711, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13120.2, 11914.5, 13128.8], "seq_len": 256}, {"decode_tps": [365.552, 372.554, 373.202], "native": {"decode": {"avg_ns": 172784606, "avg_ts": 370.435965, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [175077825, 171787214, 171488781], "samples_ts": [365.552, 372.554, 373.202], "split_mode": "layer", "stddev_ns": 1991669, "stddev_ts": 4.242284, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:36Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36589078, "avg_ts": 13993.249508, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [36570987, 36585629, 36610619], "samples_ts": [14000.2, 13994.6, 13985], "split_mode": "layer", "stddev_ns": 20932, "stddev_ts": 7.663073, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14000.2, 13994.6, 13985], "seq_len": 512}, {"decode_tps": [346.201, 354.307, 351.911], "native": {"decode": {"avg_ns": 182453984, "avg_ts": 350.806477, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [184863604, 180634293, 181864056], "samples_ts": [346.201, 354.307, 351.911], "split_mode": "layer", "stddev_ns": 2175537, "stddev_ts": 4.164311, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:37Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77644451, "avg_ts": 13188.332505, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [77630818, 77736892, 77565645], "samples_ts": [13190.6, 13172.6, 13201.7], "split_mode": "layer", "stddev_ns": 87327, "stddev_ts": 14.677479, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:53:38Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13190.6, 13172.6, 13201.7], "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": 3.1126738770399243, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.13525390625, "load_per_core": 0.0354766845703125, "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:          21.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +47.5\u00b0C  ", "Tccd1:        +43.2\u00b0C  ", "Tccd2:        +32.9\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:        +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:    +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": 1.13525390625, "load_per_core": 0.0354766845703125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +38.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.8\u00b0C  ", "Tccd1:        +38.0\u00b0C  ", "Tccd2:        +31.0\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:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +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)"]}, "position": 1, "rows": [{"decode_tps": [290.86413917849063, 352.22507182089356, 352.4229074889868], "native": {"agg": "mean", "decode_ms": 194.45, "decode_ms_best": 181.6, "decode_ms_worst": 220.03, "decode_n": 64, "decode_tps": 329.141, "prefill_ms": 17.27, "prefill_ms_best": 14.83, "prefill_ms_worst": 22.05, "prefill_tps": 7410.877, "repeats": 3, "repeats_requested": 3, "rss_mb": 285.64, "samples": {"decode_ms": [220.034, 181.702, 181.6], "prefill_ms": [22.052, 14.929, 14.834], "total_ms": [242.086, 196.631, 196.435], "ttft_ms": [22.052, 14.929, 14.834]}, "seq_len": 128, "threads": 4, "total_ms": 211.72, "total_ms_best": 196.43, "total_ms_worst": 242.09, "total_tps": 906.869, "total_tps_best": 977.424, "total_tps_worst": 793.105, "ttft_ms": 17.27, "warmup": 64}, "prefill_tps": [5804.462180301107, 8573.916538281197, 8628.82567075637], "seq_len": 128}, {"decode_tps": [351.4899880273723, 351.2315054660403, 351.1158901885053], "native": {"agg": "mean", "decode_ms": 182.19, "decode_ms_best": 182.08, "decode_ms_worst": 182.28, "decode_n": 64, "decode_tps": 351.279, "prefill_ms": 23.2, "prefill_ms_best": 23.2, "prefill_ms_worst": 23.21, "prefill_tps": 11032.463, "repeats": 3, "repeats_requested": 3, "rss_mb": 285.64, "samples": {"decode_ms": [182.082, 182.216, 182.276], "prefill_ms": [23.203, 23.203, 23.207], "total_ms": [205.286, 205.419, 205.483], "ttft_ms": [23.203, 23.203, 23.207]}, "seq_len": 256, "threads": 4, "total_ms": 205.4, "total_ms_best": 205.29, "total_ms_worst": 205.48, "total_tps": 1557.968, "total_tps_best": 1558.804, "total_tps_worst": 1557.309, "ttft_ms": 23.2, "warmup": 64}, "prefill_tps": [11033.056070335733, 11033.056070335733, 11031.154393071056], "seq_len": 256}, {"decode_tps": [344.27481737296796, 344.18039354876873, 341.79087738786325], "native": {"agg": "mean", "decode_ms": 186.37, "decode_ms_best": 185.9, "decode_ms_worst": 187.25, "decode_n": 64, "decode_tps": 343.412, "prefill_ms": 38.59, "prefill_ms_best": 38.55, "prefill_ms_worst": 38.6, "prefill_tps": 13269.295, "repeats": 3, "repeats_requested": 3, "rss_mb": 285.64, "samples": {"decode_ms": [185.898, 185.949, 187.249], "prefill_ms": [38.551, 38.604, 38.601], "total_ms": [224.449, 224.553, 225.85], "ttft_ms": [38.551, 38.604, 38.601]}, "seq_len": 512, "threads": 4, "total_ms": 224.95, "total_ms_best": 224.45, "total_ms_worst": 225.85, "total_tps": 2560.563, "total_tps_best": 2566.284, "total_tps_worst": 2550.368, "ttft_ms": 38.59, "warmup": 64}, "prefill_tps": [13281.108142460636, 13262.874313542638, 13263.905080179271], "seq_len": 512}, {"decode_tps": [326.60059808734525, 326.89419864951833, 326.65227355086336], "native": {"agg": "mean", "decode_ms": 195.89, "decode_ms_best": 195.78, "decode_ms_worst": 195.96, "decode_n": 64, "decode_tps": 326.716, "prefill_ms": 85.89, "prefill_ms_best": 85.83, "prefill_ms_worst": 85.96, "prefill_tps": 11922.504, "repeats": 3, "repeats_requested": 3, "rss_mb": 285.64, "samples": {"decode_ms": [195.958, 195.782, 195.927], "prefill_ms": [85.961, 85.877, 85.826], "total_ms": [281.918, 281.659, 281.753], "ttft_ms": [85.961, 85.878, 85.826]}, "seq_len": 1024, "threads": 4, "total_ms": 281.78, "total_ms_best": 281.66, "total_ms_worst": 281.92, "total_tps": 3861.21, "total_tps_best": 3862.824, "total_tps_worst": 3859.273, "ttft_ms": 85.89, "warmup": 64}, "prefill_tps": [11912.378869487326, 11924.030881376853, 11931.116444900148], "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": 3.1592477939557284, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.126953125, "load_per_core": 0.03521728515625, "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:         +39.0\u00b0C  ", "PPT:          24.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +43.2\u00b0C  ", "Tccd1:        +43.0\u00b0C  ", "Tccd2:        +32.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +34.9\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:    +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": 1.05078125, "load_per_core": 0.0328369140625, "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:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.6\u00b0C  ", "Tccd1:        +38.0\u00b0C  ", "Tccd2:        +31.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +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:    +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)"]}, "position": 0, "rows": [{"decode_tps": [267.20999365376264, 348.48897359106996, 349.57013797096386], "native": {"agg": "mean", "decode_ms": 202.08, "decode_ms_best": 183.08, "decode_ms_worst": 239.51, "decode_n": 64, "decode_tps": 316.704, "prefill_ms": 17.3, "prefill_ms_best": 14.84, "prefill_ms_worst": 22.09, "prefill_tps": 7399.445, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [239.512, 183.65, 183.082], "prefill_ms": [22.087, 14.964, 14.845], "total_ms": [261.599, 198.615, 197.927], "ttft_ms": [22.087, 14.965, 14.845]}, "seq_len": 128, "threads": 4, "total_ms": 219.38, "total_ms_best": 197.93, "total_ms_worst": 261.6, "total_tps": 875.194, "total_tps_best": 970.057, "total_tps_worst": 733.948, "ttft_ms": 17.3, "warmup": 64}, "prefill_tps": [5795.264182550822, 8553.862603581929, 8622.431795217244], "seq_len": 128}, {"decode_tps": [348.77004065350786, 348.43584969348535, 348.81566182321586], "native": {"agg": "mean", "decode_ms": 183.55, "decode_ms_best": 183.48, "decode_ms_worst": 183.68, "decode_n": 64, "decode_tps": 348.674, "prefill_ms": 23.23, "prefill_ms_best": 23.22, "prefill_ms_worst": 23.24, "prefill_tps": 11020.849, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [183.502, 183.678, 183.478], "prefill_ms": [23.227, 23.222, 23.237], "total_ms": [206.729, 206.9, 206.715], "ttft_ms": [23.227, 23.222, 23.237]}, "seq_len": 256, "threads": 4, "total_ms": 206.78, "total_ms_best": 206.71, "total_ms_worst": 206.9, "total_tps": 1547.529, "total_tps_best": 1548.028, "total_tps_worst": 1546.639, "ttft_ms": 23.23, "warmup": 64}, "prefill_tps": [11021.655831575321, 11024.028938075962, 11016.91268236003], "seq_len": 256}, {"decode_tps": [342.0722093054331, 341.4225584286027, 341.5592178293912], "native": {"agg": "mean", "decode_ms": 187.31, "decode_ms_best": 187.1, "decode_ms_worst": 187.45, "decode_n": 64, "decode_tps": 341.684, "prefill_ms": 38.62, "prefill_ms_best": 38.61, "prefill_ms_worst": 38.64, "prefill_tps": 13256.331, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [187.095, 187.451, 187.376], "prefill_ms": [38.621, 38.64, 38.608], "total_ms": [225.716, 226.091, 225.984], "ttft_ms": [38.621, 38.64, 38.608]}, "seq_len": 512, "threads": 4, "total_ms": 225.93, "total_ms_best": 225.72, "total_ms_worst": 226.09, "total_tps": 2549.458, "total_tps_best": 2551.878, "total_tps_worst": 2547.647, "ttft_ms": 38.62, "warmup": 64}, "prefill_tps": [13257.036327386653, 13250.517598343686, 13261.500207210942], "seq_len": 512}, {"decode_tps": [323.05284942708596, 324.28214573442307, 324.4465398283475], "native": {"agg": "mean", "decode_ms": 197.58, "decode_ms_best": 197.26, "decode_ms_worst": 198.11, "decode_n": 64, "decode_tps": 323.926, "prefill_ms": 85.91, "prefill_ms_best": 85.82, "prefill_ms_worst": 86.0, "prefill_tps": 11918.91, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [198.11, 197.359, 197.259], "prefill_ms": [86.0, 85.82, 85.921], "total_ms": [284.11, 283.179, 283.181], "ttft_ms": [86.001, 85.82, 85.921]}, "seq_len": 1024, "threads": 4, "total_ms": 283.49, "total_ms_best": 283.18, "total_ms_worst": 284.11, "total_tps": 3837.879, "total_tps_best": 3842.093, "total_tps_worst": 3829.499, "ttft_ms": 85.91, "warmup": 64}, "prefill_tps": [11906.976744186046, 11931.950594267071, 11917.924605160553], "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": 3.5713732233271003, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 1.34375, "load_per_core": 0.0419921875, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.38 V  ", "vddnb:         1.19 V  ", "edge:         +40.0\u00b0C  ", "PPT:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.8\u00b0C  ", "Tccd1:        +46.6\u00b0C  ", "Tccd2:        +32.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +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": 1.025390625, "load_per_core": 0.03204345703125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +38.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.5\u00b0C  ", "Tccd1:        +38.9\u00b0C  ", "Tccd2:        +31.9\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +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:    +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [378.829, 388.152, 388.668], "native": {"decode": {"avg_ns": 166163517, "avg_ts": 385.216283, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [168941853, 164883849, 164664850], "samples_ts": [378.829, 388.152, 388.668], "split_mode": "layer", "stddev_ns": 2408634, "stddev_ts": 5.537943, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:16Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 18528931, "avg_ts": 6908.248571, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [18635809, 18512324, 18438661], "samples_ts": [6868.5, 6914.31, 6941.94], "split_mode": "layer", "stddev_ns": 99710, "stddev_ts": 37.093002, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:15Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [6868.5, 6914.31, 6941.94], "seq_len": 128}, {"decode_tps": [375.366, 385.149, 377.849], "native": {"decode": {"avg_ns": 168683272, "avg_ts": 379.454471, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [170500480, 166169374, 169379962], "samples_ts": [375.366, 385.149, 377.849], "split_mode": "layer", "stddev_ns": 2248033, "stddev_ts": 5.085661, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:17Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 25432680, "avg_ts": 10133.460276, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [26803849, 26916454, 22577738], "samples_ts": [9550.87, 9510.91, 11338.6], "split_mode": "layer", "stddev_ns": 2473098, "stddev_ts": 1043.875105, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:15Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9550.87, 9510.91, 11338.6], "seq_len": 256}, {"decode_tps": [364.426, 367.597, 368.014], "native": {"decode": {"avg_ns": 174542966, "avg_ts": 366.678916, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [175618849, 174103535, 173906516], "samples_ts": [364.426, 367.597, 368.014], "split_mode": "layer", "stddev_ns": 937120, "stddev_ts": 1.962529, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:17Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36144453, "avg_ts": 14165.415144, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [36091197, 36122089, 36220073], "samples_ts": [14186.3, 14174.2, 14135.8], "split_mode": "layer", "stddev_ns": 67285, "stddev_ts": 26.348289, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:15Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14186.3, 14174.2, 14135.8], "seq_len": 512}, {"decode_tps": [340.062, 354.156, 353.14], "native": {"decode": {"avg_ns": 183381319, "avg_ts": 349.119072, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [188201199, 180711339, 181231420], "samples_ts": [340.062, 354.156, 353.14], "split_mode": "layer", "stddev_ns": 4182252, "stddev_ts": 7.860448, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:18Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77244971, "avg_ts": 13256.566974, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [77162208, 77136877, 77435830], "samples_ts": [13270.7, 13275.1, 13223.9], "split_mode": "layer", "stddev_ns": 166238, "stddev_ts": 28.41517, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:58:16Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13270.7, 13275.1, 13223.9], "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": 3.552762273931876, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 1.4931640625, "load_per_core": 0.046661376953125, "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:          21.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +45.4\u00b0C  ", "Tccd1:        +38.8\u00b0C  ", "Tccd2:        +33.6\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +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": 1.1875, "load_per_core": 0.037109375, "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:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.5\u00b0C  ", "Tccd1:        +37.2\u00b0C  ", "Tccd2:        +31.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [313.291, 386.812, 385.925], "native": {"decode": {"avg_ns": 178524361, "avg_ts": 362.009462, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 128, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [204282759, 165454844, 165835480], "samples_ts": [313.291, 386.812, 385.925], "split_mode": "layer", "stddev_ns": 22308238, "stddev_ts": 42.193549, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:49Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 13211133, "avg_ts": 9689.370688, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 128, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [13327151, 13226267, 13079981], "samples_ts": [9604.45, 9677.71, 9785.95], "split_mode": "layer", "stddev_ns": 124278, "stddev_ts": 91.306782, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:52Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9604.45, 9677.71, 9785.95], "seq_len": 128}, {"decode_tps": [375.035, 387.152, 380.381], "native": {"decode": {"avg_ns": 168070996, "avg_ts": 380.855888, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 256, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [170650687, 165309818, 168252485], "samples_ts": [375.035, 387.152, 380.381], "split_mode": "layer", "stddev_ns": 2675118, "stddev_ts": 6.072322, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:50Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 19486766, "avg_ts": 13137.157195, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 256, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [19517387, 19441944, 19500967], "samples_ts": [13116.5, 13167.4, 13127.6], "split_mode": "layer", "stddev_ns": 39675, "stddev_ts": 26.773237, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:52Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13116.5, 13167.4, 13127.6], "seq_len": 256}, {"decode_tps": [366.784, 370.86, 372.223], "native": {"decode": {"avg_ns": 173000558, "avg_ts": 369.955483, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 512, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [174489672, 172572030, 171939972], "samples_ts": [366.784, 370.86, 372.223], "split_mode": "layer", "stddev_ns": 1327768, "stddev_ts": 2.830002, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:50Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36008814, "avg_ts": 14218.770589, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 512, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [35946492, 36006024, 36073927], "samples_ts": [14243.4, 14219.8, 14193.1], "split_mode": "layer", "stddev_ns": 64045, "stddev_ts": 25.175282, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:52Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14243.4, 14219.8, 14193.1], "seq_len": 512}, {"decode_tps": [342.683, 351.562, 350.876], "native": {"decode": {"avg_ns": 183735691, "avg_ts": 348.373519, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 1024, "n_gen": 64, "n_gpu_layers": 99, "n_prompt": 0, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [186761418, 182044846, 182400811], "samples_ts": [342.683, 351.562, 350.876], "split_mode": "layer", "stddev_ns": 2626463, "stddev_ts": 4.939888, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:51Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77516147, "avg_ts": 13210.174333, "backends": "Vulkan", "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": "NVIDIA GeForce RTX 2080 Ti", "lazy_mode": "auto", "load_mode": "auto", "main_gpu": 0, "model_filename": "/home/germar/gguf-campaign/bitnet_b1_58-large-TQ2_0.gguf", "model_n_params": 728842752, "model_size": 216341976, "model_type": "bitnet ?B TQ2_0 - 2.06 bpw ternary", "n_batch": 2048, "n_cpu_moe": 0, "n_depth": 0, "n_gen": 0, "n_gpu_layers": 99, "n_prompt": 1024, "n_threads": 4, "n_ubatch": 512, "no_host": false, "no_kv_offload": false, "no_op_offload": 0, "poll": 100, "repack": true, "samples_ns": [77518084, 77641729, 77388629], "samples_ts": [13209.8, 13188.8, 13231.9], "split_mode": "layer", "stddev_ns": 126866, "stddev_ts": 21.569174, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T18:59:52Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13209.8, 13188.8, 13231.9], "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": 3.1194907119497657, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 1.22119140625, "load_per_core": 0.0381622314453125, "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:          22.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +43.6\u00b0C  ", "Tccd1:        +40.6\u00b0C  ", "Tccd2:        +32.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +34.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 1.22119140625, "load_per_core": 0.0381622314453125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.37 V  ", "vddnb:         1.19 V  ", "edge:         +38.0\u00b0C  ", "PPT:          14.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +44.0\u00b0C  ", "Tccd1:        +36.2\u00b0C  ", "Tccd2:        +31.4\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +32.0\u00b0C  ", "", "ucsi_source_psy_2_00081-i2c-2-08", "Adapter: NVIDIA GPU I2C adapter", "in0:           0.00 V  (min =  +0.00 V, max =  +0.00 V)", "curr1:         0.00 A  (max =  +0.00 A)", "", "nvme-pci-0e00", "Adapter: PCI adapter", "Composite:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +36.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "position": 1, "rows": [{"decode_tps": [280.6279049372972, 351.337553043736, 351.8068580349388], "native": {"agg": "mean", "decode_ms": 197.38, "decode_ms_best": 181.92, "decode_ms_worst": 228.06, "decode_n": 64, "decode_tps": 324.248, "prefill_ms": 17.28, "prefill_ms_best": 14.85, "prefill_ms_worst": 22.07, "prefill_tps": 7406.671, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [228.06, 182.161, 181.918], "prefill_ms": [22.067, 14.931, 14.847], "total_ms": [250.127, 197.092, 196.765], "ttft_ms": [22.067, 14.931, 14.847]}, "seq_len": 128, "threads": 4, "total_ms": 214.66, "total_ms_best": 196.77, "total_ms_worst": 250.13, "total_tps": 894.432, "total_tps_best": 975.782, "total_tps_worst": 767.61, "ttft_ms": 17.28, "warmup": 64}, "prefill_tps": [5800.516608510446, 8572.768066438954, 8621.270290294337], "seq_len": 128}, {"decode_tps": [350.854110475188, 350.27803318884366, 350.44106293154897], "native": {"agg": "mean", "decode_ms": 182.58, "decode_ms_best": 182.41, "decode_ms_worst": 182.71, "decode_n": 64, "decode_tps": 350.524, "prefill_ms": 23.22, "prefill_ms_best": 23.21, "prefill_ms_worst": 23.22, "prefill_tps": 11027.173, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [182.412, 182.712, 182.627], "prefill_ms": [23.225, 23.212, 23.209], "total_ms": [205.637, 205.924, 205.837], "ttft_ms": [23.225, 23.212, 23.21]}, "seq_len": 256, "threads": 4, "total_ms": 205.8, "total_ms_best": 205.64, "total_ms_worst": 205.92, "total_tps": 1554.914, "total_tps_best": 1556.141, "total_tps_worst": 1553.972, "ttft_ms": 23.22, "warmup": 64}, "prefill_tps": [11022.604951560817, 11028.77821816302, 11030.203800249903], "seq_len": 256}, {"decode_tps": [343.4176495208251, 343.764435420628, 344.3859705765237], "native": {"agg": "mean", "decode_ms": 186.12, "decode_ms_best": 185.84, "decode_ms_worst": 186.36, "decode_n": 64, "decode_tps": 343.856, "prefill_ms": 38.6, "prefill_ms_best": 38.6, "prefill_ms_worst": 38.61, "prefill_tps": 13263.502, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [186.362, 186.174, 185.838], "prefill_ms": [38.61, 38.597, 38.599], "total_ms": [224.973, 224.771, 224.437], "ttft_ms": [38.611, 38.597, 38.599]}, "seq_len": 512, "threads": 4, "total_ms": 224.73, "total_ms_best": 224.44, "total_ms_worst": 224.97, "total_tps": 2563.112, "total_tps_best": 2566.424, "total_tps_worst": 2560.31, "ttft_ms": 38.6, "warmup": 64}, "prefill_tps": [13260.813260813262, 13265.279684949606, 13264.592346951995], "seq_len": 512}, {"decode_tps": [326.55227132411846, 327.0094475073194, 326.84912338043705], "native": {"agg": "mean", "decode_ms": 195.84, "decode_ms_best": 195.71, "decode_ms_worst": 195.99, "decode_n": 64, "decode_tps": 326.804, "prefill_ms": 85.88, "prefill_ms_best": 85.8, "prefill_ms_worst": 85.93, "prefill_tps": 11923.951, "repeats": 3, "repeats_requested": 3, "rss_mb": 286.77, "samples": {"decode_ms": [195.987, 195.713, 195.809], "prefill_ms": [85.925, 85.799, 85.908], "total_ms": [281.912, 281.512, 281.717], "ttft_ms": [85.925, 85.799, 85.909]}, "seq_len": 1024, "threads": 4, "total_ms": 281.71, "total_ms_best": 281.51, "total_ms_worst": 281.91, "total_tps": 3862.075, "total_tps_best": 3864.838, "total_tps_worst": 3859.364, "ttft_ms": 85.88, "warmup": 64}, "prefill_tps": [11917.369799243526, 11934.871035792958, 11919.728081203148], "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": 0.40718269497110576, "mad_tps": 1.3253772517402354, "median_tps": 325.499405576233, "samples": 12, "vs_llama_pct": -6.610200716353287}, "prefill": {"mad_pct": 0.05932642321409355, "mad_tps": 7.071013358508935, "median_tps": 11918.82634318185, "samples": 12, "vs_llama_pct": -9.820975927079056}}, "128": {"decode": {"mad_pct": 0.9118773423655402, "mad_tps": 3.1809474862870957, "median_tps": 348.8350174416291, "samples": 12, "vs_llama_pct": -9.18202582334473}, "prefill": {"mad_pct": 0.6667584066632688, "mad_tps": 57.10615653811874, "median_tps": 8564.744886217672, "samples": 12, "vs_llama_pct": 3.4390103643501035}}, "256": {"decode": {"mad_pct": 0.3457160305589575, "mad_tps": 1.2091303908513282, "median_tps": 349.74669496707827, "samples": 12, "vs_llama_pct": -7.746542614489471}, "prefill": {"mad_pct": 0.06028638184135548, "mad_tps": 6.645558944936056, "median_tps": 11023.31694481839, "samples": 12, "vs_llama_pct": -5.188409761981061}}, "512": {"decode": {"mad_pct": 0.29863219807239394, "mad_tps": 1.022130854994316, "median_tps": 342.2708139282867, "samples": 12, "vs_llama_pct": -6.786582461069468}, "prefill": {"mad_pct": 0.03107124595609832, "mad_tps": 4.120406625448595, "median_tps": 13261.156734012102, "samples": 12, "vs_llama_pct": -6.481502554524066}}}, "llama.cpp": {"1024": {"decode": {"mad_pct": 1.3984107924949454, "mad_tps": 4.873999999999995, "median_tps": 348.5385, "samples": 12}, "prefill": {"mad_pct": 0.21260739132244344, "mad_tps": 28.100000000000364, "median_tps": 13216.849999999999, "samples": 12}}, "128": {"decode": {"mad_pct": 0.9315197596481165, "mad_tps": 3.578000000000003, "median_tps": 384.1035, "samples": 12}, "prefill": {"mad_pct": 16.454538921823023, "mad_tps": 1362.4350000000004, "median_tps": 8279.995, "samples": 12}}, "256": {"decode": {"mad_pct": 1.032536301649896, "mad_tps": 3.914500000000004, "median_tps": 379.115, "samples": 12}, "prefill": {"mad_pct": 13.086857236239473, "mad_tps": 1521.5500000000002, "median_tps": 11626.55, "samples": 12}}, "512": {"decode": {"mad_pct": 0.8761119909148012, "mad_tps": 3.217000000000013, "median_tps": 367.1905, "samples": 12}, "prefill": {"mad_pct": 0.2961865975564606, "mad_tps": 42.0, "median_tps": 14180.25, "samples": 12}}}}}