{"created_utc": "2026-10-09T16:08:29.693154+00:00", "engines": [{"binary": "/tmp/claude-1000/-home-germar-workplace-geistlib/108234fc-8402-4734-b735-41f3bce6f71b/scratchpad/wt/p6-vk/bin/linux/release/tests/bench_perf_sweep", "binary_sha256": "a2b0b08be83d014e487042cc259624fae7c778d32028bcba0c24b54abe0e464c", "commit": "8489cb9537749f11e775267fc6ab5ce65b4b845f", "kind": "geist", "label": "geist", "linked_libraries": ["linux-vdso.so.1 (0x00007c1b86e46000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x00007c1b86a55000)", "\tlibmvec.so.1 => /usr/lib/x86_64-linux-gnu/libmvec.so.1 (0x00007c1b8695a000)", "\tlibopenblas.so.0 => /usr/lib/x86_64-linux-gnu/libopenblas.so.0 (0x00007c1b84480000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007c1b84421000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x00007c1b84200000)", "\t/lib64/ld-linux-x86-64.so.2 (0x00007c1b86e48000)", "\tlibgfortran.so.5 => /usr/lib/x86_64-linux-gnu/libgfortran.so.5 (0x00007c1b83e00000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007c1b841d2000)"]}, {"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 (0x00007d8d08900000)", "\tlibllama-bench-impl.so => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama-bench-impl.so (0x00007d8d0888d000)", "\tlibstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007d8d08400000)", "\tlibgcc_s.so.1 => /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007d8d08842000)", "\tlibc.so.6 => /usr/lib/x86_64-linux-gnu/libc.so.6 (0x00007d8d08000000)", "\tlibllama-common.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama-common.so.0 (0x00007d8d07a00000)", "\tlibllama.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libllama.so.0 (0x00007d8d07400000)", "\tlibggml.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml.so.0 (0x00007d8d08832000)", "\tlibggml-base.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-base.so.0 (0x00007d8d08757000)", "\tlibm.so.6 => /usr/lib/x86_64-linux-gnu/libm.so.6 (0x00007d8d082da000)", "\t/lib64/ld-linux-x86-64.so.2 (0x00007d8d08902000)", "\tlibssl.so.3 => /usr/lib/x86_64-linux-gnu/libssl.so.3 (0x00007d8d078ef000)", "\tlibcrypto.so.3 => /usr/lib/x86_64-linux-gnu/libcrypto.so.3 (0x00007d8d06c00000)", "\tlibggml-cpu.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-cpu.so.0 (0x00007d8d06800000)", "\tlibggml-vulkan.so.0 => /home/germar/llama.cpp-v0.6.0/build-vulkan/bin/libggml-vulkan.so.0 (0x00007d8d04600000)", "\tlibgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007d8d086f6000)", "\tlibz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1 (0x00007d8d086d8000)", "\tlibzstd.so.1 => /usr/lib/x86_64-linux-gnu/libzstd.so.1 (0x00007d8d07f3d000)", "\tlibvulkan.so.1 => /usr/lib/x86_64-linux-gnu/libvulkan.so.1 (0x00007d8d0824a000)"]}], "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": "53aacaacb9d4b51c2c860720c016b8b5ea5322ec", "path": "tools/bench_cross_engine.py", "sha256": "1b2af101ff1c7bc98722fb96a3f43dff448df3808c9012cc7d69475f04c067e5"}, "schema": "geist.benchmark.cross-engine-run.v1", "system": {"compiler": "cc (Ubuntu 15.2.0-16ubuntu1) 15.2.0", "cpu": "AMD Ryzen 9 9950X 16-Core Processor", "logical_cores": 32, "memory_bytes": "63813259264", "os": "Linux 7.0.0-34-generic", "performance_cores": "16", "product": "Ubuntu 26.04.1 LTS"}}
{"cycle": 0, "elapsed_seconds": 3.637237231945619, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 0.30517578125, "load_per_core": 0.0095367431640625, "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:         +37.0\u00b0C  ", "PPT:          22.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.5\u00b0C  ", "Tccd1:        +37.5\u00b0C  ", "Tccd2:        +31.5\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": 0.0703125, "load_per_core": 0.002197265625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +35.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.4\u00b0C  ", "Tccd1:        +25.9\u00b0C  ", "Tccd2:        +25.9\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:        +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": [324.22793223636216, 348.21972664751456, 352.20762639576026], "native": {"agg": "mean", "decode_ms": 187.63, "decode_ms_best": 181.71, "decode_ms_worst": 197.39, "decode_n": 64, "decode_tps": 341.093, "prefill_ms": 18.21, "prefill_ms_best": 15.78, "prefill_ms_worst": 22.89, "prefill_tps": 7027.453, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.04, "samples": {"decode_ms": [197.392, 183.792, 181.711], "prefill_ms": [22.887, 15.975, 15.781], "total_ms": [220.28, 199.767, 197.492], "ttft_ms": [22.887, 15.975, 15.781]}, "seq_len": 128, "threads": 4, "total_ms": 205.85, "total_ms_best": 197.49, "total_ms_worst": 220.28, "total_tps": 932.735, "total_tps_best": 972.19, "total_tps_worst": 871.62, "ttft_ms": 18.21, "warmup": 64}, "prefill_tps": [5592.694542753528, 8012.519561815337, 8111.019580508206], "seq_len": 128}, {"decode_tps": [351.090850243295, 351.1602003807894, 350.754121360926], "native": {"agg": "mean", "decode_ms": 182.34, "decode_ms_best": 182.25, "decode_ms_worst": 182.46, "decode_n": 64, "decode_tps": 351.002, "prefill_ms": 34.61, "prefill_ms_best": 34.58, "prefill_ms_worst": 34.64, "prefill_tps": 7396.976, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.04, "samples": {"decode_ms": [182.289, 182.253, 182.464], "prefill_ms": [34.641, 34.601, 34.585], "total_ms": [216.93, 216.854, 217.048], "ttft_ms": [34.641, 34.601, 34.585]}, "seq_len": 256, "threads": 4, "total_ms": 216.94, "total_ms_best": 216.85, "total_ms_worst": 217.05, "total_tps": 1475.037, "total_tps_best": 1475.65, "total_tps_worst": 1474.327, "ttft_ms": 34.61, "warmup": 64}, "prefill_tps": [7390.086891256026, 7398.630097396029, 7402.05291311262], "seq_len": 256}, {"decode_tps": [344.1822444984619, 343.91776066548084, 344.395236583384], "native": {"agg": "mean", "decode_ms": 185.96, "decode_ms_best": 185.83, "decode_ms_worst": 186.09, "decode_n": 64, "decode_tps": 344.165, "prefill_ms": 82.48, "prefill_ms_best": 82.44, "prefill_ms_worst": 82.51, "prefill_tps": 6207.777, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.04, "samples": {"decode_ms": [185.948, 186.091, 185.833], "prefill_ms": [82.509, 82.481, 82.442], "total_ms": [268.457, 268.571, 268.275], "ttft_ms": [82.509, 82.481, 82.442]}, "seq_len": 512, "threads": 4, "total_ms": 268.43, "total_ms_best": 268.28, "total_ms_worst": 268.57, "total_tps": 2145.775, "total_tps_best": 2147.049, "total_tps_worst": 2144.683, "ttft_ms": 82.48, "warmup": 64}, "prefill_tps": [6205.383655116411, 6207.490209866515, 6210.4267242424985], "seq_len": 512}, {"decode_tps": [326.7089686716728, 326.792379610198, 324.61439360509644], "native": {"agg": "mean", "decode_ms": 196.3, "decode_ms_best": 195.84, "decode_ms_worst": 197.16, "decode_n": 64, "decode_tps": 326.035, "prefill_ms": 218.16, "prefill_ms_best": 217.8, "prefill_ms_worst": 218.71, "prefill_tps": 4693.853, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.04, "samples": {"decode_ms": [195.893, 195.843, 197.157], "prefill_ms": [217.97, 217.796, 218.707], "total_ms": [413.864, 413.639, 415.864], "ttft_ms": [217.971, 217.796, 218.707]}, "seq_len": 1024, "threads": 4, "total_ms": 414.46, "total_ms_best": 413.64, "total_ms_worst": 415.86, "total_tps": 2625.132, "total_tps_best": 2630.314, "total_tps_worst": 2616.241, "ttft_ms": 218.16, "warmup": 64}, "prefill_tps": [4697.894205624627, 4701.647413175632, 4682.0632169980845], "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": 4.542901776963845, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.14111328125, "load_per_core": 0.0044097900390625, "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:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +49.8\u00b0C  ", "Tccd1:        +38.4\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:     +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": 0.06591796875, "load_per_core": 0.0020599365234375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:          13.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.1\u00b0C  ", "Tccd1:        +26.5\u00b0C  ", "Tccd2:        +26.8\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:     +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [378.77, 385.821, 384.408], "native": {"decode": {"avg_ns": 167112557, "avg_ts": 382.99975, "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": [168968020, 165879849, 166489804], "samples_ts": [378.77, 385.821, 384.408], "split_mode": "layer", "stddev_ns": 1635665, "stddev_ts": 3.730714, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:35Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 18159113, "avg_ts": 7048.968923, "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": [18275159, 18141192, 18060989], "samples_ts": [7004.04, 7055.77, 7087.1], "split_mode": "layer", "stddev_ns": 108287, "stddev_ts": 41.942595, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:33Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [7004.04, 7055.77, 7087.1], "seq_len": 128}, {"decode_tps": [372.68, 386.356, 378.361], "native": {"decode": {"avg_ns": 168843331, "avg_ts": 379.132378, "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": [171729148, 165650409, 169150438], "samples_ts": [372.68, 386.356, 378.361], "split_mode": "layer", "stddev_ns": 3051039, "stddev_ts": 6.870485, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:36Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 26796765, "avg_ts": 9553.540456, "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": [26720498, 26723379, 26946419], "samples_ts": [9580.66, 9579.63, 9500.33], "split_mode": "layer", "stddev_ns": 129715, "stddev_ts": 46.080414, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:34Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9580.66, 9579.63, 9500.33], "seq_len": 256}, {"decode_tps": [366.463, 368.453, 371.177], "native": {"decode": {"avg_ns": 173588817, "avg_ts": 368.69746, "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": [174642472, 173699422, 172424558], "samples_ts": [366.463, 368.453, 371.177], "split_mode": "layer", "stddev_ns": 1113164, "stddev_ts": 2.366446, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:36Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36510205, "avg_ts": 14023.680739, "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": [36701324, 36457383, 36371910], "samples_ts": [13950.5, 14043.8, 14076.8], "split_mode": "layer", "stddev_ns": 171155, "stddev_ts": 65.530939, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:34Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13950.5, 14043.8, 14076.8], "seq_len": 512}, {"decode_tps": [345.586, 352.359, 350.144], "native": {"decode": {"avg_ns": 183202533, "avg_ts": 349.362973, "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": [185192541, 181633191, 182781868], "samples_ts": [345.586, 352.359, 350.144], "split_mode": "layer", "stddev_ns": 1816630, "stddev_ts": 3.453036, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:37Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77663779, "avg_ts": 13185.237344, "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": [78088668, 77477931, 77424739], "samples_ts": [13113.3, 13216.7, 13225.7], "split_mode": "layer", "stddev_ns": 369029, "stddev_ts": 62.466212, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:11:34Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13113.3, 13216.7, 13225.7], "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": 3.5450217141769826, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.3251953125, "load_per_core": 0.010162353515625, "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:         +42.9\u00b0C  ", "Tccd1:        +34.4\u00b0C  ", "Tccd2:        +29.6\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:        +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": 0.09228515625, "load_per_core": 0.0028839111328125, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.07 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:           9.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.2\u00b0C  ", "Tccd1:        +25.9\u00b0C  ", "Tccd2:        +26.8\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:     +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": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [326.933, 369.467, 385.787], "native": {"decode": {"avg_ns": 178291910, "avg_ts": 360.729093, "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": [195758508, 173222518, 165894706], "samples_ts": [326.933, 369.467, 385.787], "split_mode": "layer", "stddev_ns": 15563935, "stddev_ts": 30.384124, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:08Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 13084757, "avg_ts": 9782.837736, "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": [13208282, 13050316, 12995673], "samples_ts": [9690.89, 9808.19, 9849.43], "split_mode": "layer", "stddev_ns": 110409, "stddev_ts": 82.256117, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:11Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9690.89, 9808.19, 9849.43], "seq_len": 128}, {"decode_tps": [373.306, 382.606, 379.019], "native": {"decode": {"avg_ns": 169190673, "avg_ts": 378.310317, "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": [171441142, 167274015, 168856863], "samples_ts": [373.306, 382.606, 379.019], "split_mode": "layer", "stddev_ns": 2103563, "stddev_ts": 4.690251, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:09Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 19392799, "avg_ts": 13200.787678, "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": [19417797, 19374765, 19385836], "samples_ts": [13183.8, 13213.1, 13205.5], "split_mode": "layer", "stddev_ns": 22774, "stddev_ts": 15.203059, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:11Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13183.8, 13213.1, 13205.5], "seq_len": 256}, {"decode_tps": [364.059, 368.459, 369.727], "native": {"decode": {"avg_ns": 174197546, "avg_ts": 367.415089, "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": [175795568, 173696199, 173100873], "samples_ts": [364.059, 368.459, 369.727], "split_mode": "layer", "stddev_ns": 1415699, "stddev_ts": 2.974519, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:09Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36475330, "avg_ts": 14036.885992, "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": [36489605, 36476774, 36459613], "samples_ts": [14031.4, 14036.3, 14042.9], "split_mode": "layer", "stddev_ns": 17302, "stddev_ts": 5.791299, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:11Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14031.4, 14036.3, 14042.9], "seq_len": 512}, {"decode_tps": [346.248, 351.694, 349.418], "native": {"decode": {"avg_ns": 183325578, "avg_ts": 349.119983, "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": [184838673, 181976171, 183161891], "samples_ts": [346.248, 351.694, 349.418], "split_mode": "layer", "stddev_ns": 1438317, "stddev_ts": 2.735431, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:10Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77913295, "avg_ts": 13142.85404, "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": [77726480, 78039004, 77974401], "samples_ts": [13174.4, 13121.6, 13132.5], "split_mode": "layer", "stddev_ns": 164979, "stddev_ts": 27.857986, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:13:11Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13174.4, 13121.6, 13132.5], "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": 3.6569382487796247, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 0.3681640625, "load_per_core": 0.011505126953125, "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:         +37.0\u00b0C  ", "PPT:          18.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.5\u00b0C  ", "Tccd1:        +39.6\u00b0C  ", "Tccd2:        +29.6\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:        +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": 0.138671875, "load_per_core": 0.00433349609375, "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:          11.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.6\u00b0C  ", "Tccd1:        +25.4\u00b0C  ", "Tccd2:        +25.4\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:    +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": [311.20685044079534, 349.94094746511524, 349.5911422344336], "native": {"agg": "mean", "decode_ms": 190.54, "decode_ms_best": 182.89, "decode_ms_worst": 205.65, "decode_n": 64, "decode_tps": 335.894, "prefill_ms": 18.22, "prefill_ms_best": 15.8, "prefill_ms_worst": 23.07, "prefill_tps": 7024.9, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.94, "samples": {"decode_ms": [205.651, 182.888, 183.071], "prefill_ms": [23.066, 15.798, 15.799], "total_ms": [228.717, 198.686, 198.869], "ttft_ms": [23.066, 15.798, 15.799]}, "seq_len": 128, "threads": 4, "total_ms": 208.76, "total_ms_best": 198.69, "total_ms_worst": 228.72, "total_tps": 919.728, "total_tps_best": 966.348, "total_tps_worst": 839.466, "ttft_ms": 18.22, "warmup": 64}, "prefill_tps": [5549.293332177231, 8102.291429294847, 8101.778593581873], "seq_len": 128}, {"decode_tps": [349.34688508125043, 348.81946412609824, 348.0530780944094], "native": {"agg": "mean", "decode_ms": 183.52, "decode_ms_best": 183.2, "decode_ms_worst": 183.88, "decode_n": 64, "decode_tps": 348.739, "prefill_ms": 34.69, "prefill_ms_best": 34.59, "prefill_ms_worst": 34.88, "prefill_tps": 7378.846, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.94, "samples": {"decode_ms": [183.199, 183.476, 183.88], "prefill_ms": [34.876, 34.595, 34.61], "total_ms": [218.075, 218.071, 218.49], "ttft_ms": [34.876, 34.595, 34.61]}, "seq_len": 256, "threads": 4, "total_ms": 218.21, "total_ms_best": 218.07, "total_ms_worst": 218.49, "total_tps": 1466.464, "total_tps_best": 1467.414, "total_tps_worst": 1464.597, "ttft_ms": 34.69, "warmup": 64}, "prefill_tps": [7340.291317811676, 7399.913282266224, 7396.706154290668], "seq_len": 256}, {"decode_tps": [341.86025393806983, 341.58656283858437, 342.1380420081365], "native": {"agg": "mean", "decode_ms": 187.21, "decode_ms_best": 187.06, "decode_ms_worst": 187.36, "decode_n": 64, "decode_tps": 341.861, "prefill_ms": 82.62, "prefill_ms_best": 82.48, "prefill_ms_worst": 82.84, "prefill_tps": 6197.344, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.94, "samples": {"decode_ms": [187.211, 187.361, 187.059], "prefill_ms": [82.529, 82.484, 82.836], "total_ms": [269.74, 269.845, 269.895], "ttft_ms": [82.529, 82.484, 82.836]}, "seq_len": 512, "threads": 4, "total_ms": 269.83, "total_ms_best": 269.74, "total_ms_worst": 269.89, "total_tps": 2134.705, "total_tps_best": 2135.39, "total_tps_worst": 2134.167, "ttft_ms": 82.62, "warmup": 64}, "prefill_tps": [6203.879848295751, 6207.26443916396, 6180.8875368197405], "seq_len": 512}, {"decode_tps": [322.6814831247668, 324.53867334675436, 324.5748829755402], "native": {"agg": "mean", "decode_ms": 197.57, "decode_ms_best": 197.18, "decode_ms_worst": 198.34, "decode_n": 64, "decode_tps": 323.929, "prefill_ms": 217.9, "prefill_ms_best": 217.82, "prefill_ms_worst": 218.01, "prefill_tps": 4699.477, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.94, "samples": {"decode_ms": [198.338, 197.203, 197.181], "prefill_ms": [218.011, 217.817, 217.862], "total_ms": [416.349, 415.02, 415.043], "ttft_ms": [218.011, 217.818, 217.862]}, "seq_len": 1024, "threads": 4, "total_ms": 415.47, "total_ms_best": 415.02, "total_ms_worst": 416.35, "total_tps": 2618.718, "total_tps_best": 2621.56, "total_tps_worst": 2613.195, "ttft_ms": 217.9, "warmup": 64}, "prefill_tps": [4697.010701294889, 4701.194121670944, 4700.223076993693], "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": 3.682917193043977, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 0.1044921875, "load_per_core": 0.003265380859375, "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:         +41.4\u00b0C  ", "Tccd1:        +36.1\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:    +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": 0.1044921875, "load_per_core": 0.003265380859375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.34 V  ", "vddnb:         1.19 V  ", "edge:         +35.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.5\u00b0C  ", "Tccd1:        +27.2\u00b0C  ", "Tccd2:        +27.2\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +33.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "", "mt7921_phy0-pci-0b00", "Adapter: PCI adapter", "temp1:        +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": [271.6860667498705, 349.29540567386726, 349.9562554680665], "native": {"agg": "mean", "decode_ms": 200.56, "decode_ms_best": 182.88, "decode_ms_worst": 235.57, "decode_n": 64, "decode_tps": 319.111, "prefill_ms": 18.29, "prefill_ms_best": 15.77, "prefill_ms_worst": 23.19, "prefill_tps": 6999.94, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.34, "samples": {"decode_ms": [235.566, 183.226, 182.88], "prefill_ms": [23.192, 15.891, 15.775], "total_ms": [258.758, 199.117, 198.655], "ttft_ms": [23.192, 15.891, 15.775]}, "seq_len": 128, "threads": 4, "total_ms": 218.84, "total_ms_best": 198.66, "total_ms_worst": 258.76, "total_tps": 877.34, "total_tps_best": 966.499, "total_tps_worst": 742.007, "ttft_ms": 18.29, "warmup": 64}, "prefill_tps": [5519.144532597447, 8054.873827952929, 8114.104595879556], "seq_len": 128}, {"decode_tps": [347.64848392668966, 348.6598387448246, 347.8714614949776], "native": {"agg": "mean", "decode_ms": 183.88, "decode_ms_best": 183.56, "decode_ms_worst": 184.09, "decode_n": 64, "decode_tps": 348.059, "prefill_ms": 34.66, "prefill_ms_best": 34.64, "prefill_ms_worst": 34.69, "prefill_tps": 7386.333, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.34, "samples": {"decode_ms": [184.094, 183.56, 183.976], "prefill_ms": [34.694, 34.639, 34.643], "total_ms": [218.787, 218.199, 218.62], "ttft_ms": [34.694, 34.639, 34.644]}, "seq_len": 256, "threads": 4, "total_ms": 218.54, "total_ms_best": 218.2, "total_ms_worst": 218.79, "total_tps": 1464.294, "total_tps_best": 1466.551, "total_tps_worst": 1462.608, "ttft_ms": 34.66, "warmup": 64}, "prefill_tps": [7378.7974865971055, 7390.5135829556275, 7389.660248823716], "seq_len": 256}, {"decode_tps": [342.31371982691763, 344.1470798582544, 341.637618572275], "native": {"agg": "mean", "decode_ms": 186.75, "decode_ms_best": 185.97, "decode_ms_worst": 187.33, "decode_n": 64, "decode_tps": 342.696, "prefill_ms": 82.6, "prefill_ms_best": 82.52, "prefill_ms_worst": 82.72, "prefill_tps": 6198.358, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.34, "samples": {"decode_ms": [186.963, 185.967, 187.333], "prefill_ms": [82.565, 82.523, 82.72], "total_ms": [269.528, 268.489, 270.053], "ttft_ms": [82.565, 82.523, 82.72]}, "seq_len": 512, "threads": 4, "total_ms": 269.36, "total_ms_best": 268.49, "total_ms_worst": 270.05, "total_tps": 2138.428, "total_tps_best": 2145.336, "total_tps_worst": 2132.913, "ttft_ms": 82.6, "warmup": 64}, "prefill_tps": [6201.174831950584, 6204.3309138058485, 6189.555125725338], "seq_len": 512}, {"decode_tps": [324.2049380464626, 324.142541682705, 326.2093958500048], "native": {"agg": "mean", "decode_ms": 197.01, "decode_ms_best": 196.19, "decode_ms_worst": 197.44, "decode_n": 64, "decode_tps": 324.85, "prefill_ms": 218.39, "prefill_ms_best": 217.89, "prefill_ms_worst": 218.93, "prefill_tps": 4688.865, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.34, "samples": {"decode_ms": [197.406, 197.444, 196.193], "prefill_ms": [218.357, 217.887, 218.926], "total_ms": [415.763, 415.331, 415.118], "ttft_ms": [218.357, 217.887, 218.926]}, "seq_len": 1024, "threads": 4, "total_ms": 415.4, "total_ms_best": 415.12, "total_ms_worst": 415.76, "total_tps": 2619.138, "total_tps_best": 2620.94, "total_tps_worst": 2616.878, "ttft_ms": 218.39, "warmup": 64}, "prefill_tps": [4689.568001025843, 4699.683781042467, 4677.379571179303], "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": 3.553899688879028, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.0185546875, "load_per_core": 0.000579833984375, "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:          23.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.4\u00b0C  ", "Tccd1:        +40.6\u00b0C  ", "Tccd2:        +31.0\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:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.0205078125, "load_per_core": 0.000640869140625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +35.0\u00b0C  ", "PPT:          12.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.5\u00b0C  ", "Tccd1:        +28.6\u00b0C  ", "Tccd2:        +26.8\u00b0C  ", "", "nvme-pci-0200", "Adapter: PCI adapter", "Composite:    +30.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +30.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +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)"]}, "phase_order": ["prefill", "decode"], "position": 1, "rows": [{"decode_tps": [384.123, 386.639, 388.316], "native": {"decode": {"avg_ns": 165652061, "avg_ts": 386.359664, "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": [166613256, 165528891, 164814037], "samples_ts": [384.123, 386.639, 388.316], "split_mode": "layer", "stddev_ns": 906002, "stddev_ts": 2.110626, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:50Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 18542807, "avg_ts": 6903.064177, "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": [18628765, 18557001, 18442656], "samples_ts": [6871.09, 6897.67, 6940.43], "split_mode": "layer", "stddev_ns": 93961, "stddev_ts": 34.982628, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:49Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [6871.09, 6897.67, 6940.43], "seq_len": 128}, {"decode_tps": [377.167, 385.687, 379.417], "native": {"decode": {"avg_ns": 168101034, "avg_ts": 380.757388, "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": [169685935, 165937471, 168679696], "samples_ts": [377.167, 385.687, 379.417], "split_mode": "layer", "stddev_ns": 1940072, "stddev_ts": 4.41528, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:51Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 26832191, "avg_ts": 9540.918809, "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": [26904962, 26904692, 26686921], "samples_ts": [9514.97, 9515.07, 9592.71], "split_mode": "layer", "stddev_ns": 126021, "stddev_ts": 44.855954, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:50Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9514.97, 9515.07, 9592.71], "seq_len": 256}, {"decode_tps": [369.511, 366.963, 372.817], "native": {"decode": {"avg_ns": 173090784, "avg_ts": 369.763682, "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": [173202057, 174404335, 171665961], "samples_ts": [369.511, 366.963, 372.817], "split_mode": "layer", "stddev_ns": 1372637, "stddev_ts": 2.935049, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:51Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36297110, "avg_ts": 14105.850123, "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": [36385698, 36269822, 36235811], "samples_ts": [14071.5, 14116.4, 14129.7], "split_mode": "layer", "stddev_ns": 78811, "stddev_ts": 30.508175, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:50Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14071.5, 14116.4, 14129.7], "seq_len": 512}, {"decode_tps": [348.254, 354.489, 352.534], "native": {"decode": {"avg_ns": 181952828, "avg_ts": 351.758853, "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": [183774085, 180541477, 181542924], "samples_ts": [348.254, 354.489, 352.534], "split_mode": "layer", "stddev_ns": 1654937, "stddev_ts": 3.189151, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:52Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 78049117, "avg_ts": 13120.166772, "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": [78071025, 77644264, 78432063], "samples_ts": [13116.3, 13188.4, 13055.9], "split_mode": "layer", "stddev_ns": 394455, "stddev_ts": 66.320491, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:17:50Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13116.3, 13188.4, 13055.9], "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": 3.537357961758971, "engine": "llama.cpp", "kind": "run", "machine_after": {"load_1m": 0.48779296875, "load_per_core": 0.0152435302734375, "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:          18.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +42.6\u00b0C  ", "Tccd1:        +36.2\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:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.26904296875, "load_per_core": 0.0084075927734375, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.35 V  ", "vddnb:         1.19 V  ", "edge:         +35.0\u00b0C  ", "PPT:          11.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.2\u00b0C  ", "Tccd1:        +24.9\u00b0C  ", "Tccd2:        +25.4\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)"]}, "phase_order": ["decode", "prefill"], "position": 0, "rows": [{"decode_tps": [298.145, 390.212, 389.872], "native": {"decode": {"avg_ns": 180943424, "avg_ts": 359.409799, "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": [214660509, 164013514, 164156251], "samples_ts": [298.145, 390.212, 389.872], "split_mode": "layer", "stddev_ns": 29199944, "stddev_ts": 53.056972, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:23Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 13220051, "avg_ts": 9682.598826, "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": [13329713, 13176646, 13153794], "samples_ts": [9602.61, 9714.16, 9731.03], "split_mode": "layer", "stddev_ns": 95654, "stddev_ts": 69.786676, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [9602.61, 9714.16, 9731.03], "seq_len": 128}, {"decode_tps": [379.213, 387.432, 380.137], "native": {"decode": {"avg_ns": 167440439, "avg_ts": 382.260586, "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": [168770564, 165190431, 168360324], "samples_ts": [379.213, 387.432, 380.137], "split_mode": "layer", "stddev_ns": 1959416, "stddev_ts": 4.502035, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:24Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 19248891, "avg_ts": 13299.519787, "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": [19302611, 19215447, 19228617], "samples_ts": [13262.5, 13322.6, 13313.5], "split_mode": "layer", "stddev_ns": 47393, "stddev_ts": 32.422104, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13262.5, 13322.6, 13313.5], "seq_len": 256}, {"decode_tps": [369.426, 368.38, 373.249], "native": {"decode": {"avg_ns": 172814247, "avg_ts": 370.351625, "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": [173241688, 173733751, 171467302], "samples_ts": [369.426, 368.38, 373.249], "split_mode": "layer", "stddev_ns": 1192152, "stddev_ts": 2.563164, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:24Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 36081540, "avg_ts": 14190.452466, "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": [35863029, 36067330, 36314261], "samples_ts": [14276.5, 14195.7, 14099.1], "split_mode": "layer", "stddev_ns": 225951, "stddev_ts": 88.8135, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [14276.5, 14195.7, 14099.1], "seq_len": 512}, {"decode_tps": [346.631, 355.343, 353.897], "native": {"decode": {"avg_ns": 181861893, "avg_ts": 351.95697, "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": [184634447, 180107529, 180843704], "samples_ts": [346.631, 355.343, 353.897], "split_mode": "layer", "stddev_ns": 2429189, "stddev_ts": 4.668894, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:25Z", "type_k": "f16", "type_v": "f16"}, "prefill": {"avg_ns": 77815096, "avg_ts": 13159.713562, "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": [77325960, 78250994, 77868336], "samples_ts": [13242.6, 13086.1, 13150.4], "split_mode": "layer", "stddev_ns": 464976, "stddev_ts": 78.687301, "tensor_buft_overrides": "none", "tensor_split": "0.00", "test_time": "2026-10-09T16:19:26Z", "type_k": "f16", "type_v": "f16"}}, "prefill_tps": [13242.6, 13086.1, 13150.4], "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": 3.6878037550486624, "engine": "geist", "kind": "run", "machine_after": {"load_1m": 0.33837890625, "load_per_core": 0.0105743408203125, "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:         +37.0\u00b0C  ", "PPT:          20.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +41.4\u00b0C  ", "Tccd1:        +38.8\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:    +32.9\u00b0C  (low  = -273.1\u00b0C, high = +81.8\u00b0C)", "                       (crit = +84.8\u00b0C)", "Sensor 1:     +32.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)", "Sensor 2:     +35.9\u00b0C  (low  = -273.1\u00b0C, high = +65261.8\u00b0C)"]}, "machine_before": {"load_1m": 0.1064453125, "load_per_core": 0.003326416015625, "power_source": ["mains (desktop)"], "temperature_c": null, "temperature_note": "unavailable to an unprivileged process", "thermal_status": ["amdgpu-pci-0f00", "Adapter: PCI adapter", "vddgfx:        1.07 V  ", "vddnb:         1.19 V  ", "edge:         +36.0\u00b0C  ", "PPT:           8.00 mW ", "sclk:         600 MHz ", "", "k10temp-pci-00c3", "Adapter: PCI adapter", "Tctl:         +39.8\u00b0C  ", "Tccd1:        +25.4\u00b0C  ", "Tccd2:        +26.4\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:     +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)"]}, "position": 1, "rows": [{"decode_tps": [275.71589200553154, 350.604244502635, 350.2281955586687], "native": {"agg": "mean", "decode_ms": 199.13, "decode_ms_best": 182.54, "decode_ms_worst": 232.12, "decode_n": 64, "decode_tps": 321.391, "prefill_ms": 18.3, "prefill_ms_best": 15.77, "prefill_ms_worst": 23.22, "prefill_tps": 6995.97, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.98, "samples": {"decode_ms": [232.123, 182.542, 182.738], "prefill_ms": [23.224, 15.894, 15.77], "total_ms": [255.347, 198.437, 198.508], "ttft_ms": [23.225, 15.894, 15.77]}, "seq_len": 128, "threads": 4, "total_ms": 217.43, "total_ms_best": 198.44, "total_ms_worst": 255.35, "total_tps": 883.041, "total_tps_best": 967.563, "total_tps_worst": 751.918, "ttft_ms": 18.3, "warmup": 64}, "prefill_tps": [5511.539786427833, 8053.353466717, 8116.677235256817], "seq_len": 128}, {"decode_tps": [348.43584969348535, 348.4472320223006, 347.88469796541807], "native": {"agg": "mean", "decode_ms": 183.77, "decode_ms_best": 183.67, "decode_ms_worst": 183.97, "decode_n": 64, "decode_tps": 348.256, "prefill_ms": 34.62, "prefill_ms_best": 34.6, "prefill_ms_worst": 34.65, "prefill_tps": 7395.618, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.98, "samples": {"decode_ms": [183.678, 183.672, 183.969], "prefill_ms": [34.648, 34.598, 34.599], "total_ms": [218.326, 218.27, 218.568], "ttft_ms": [34.648, 34.599, 34.599]}, "seq_len": 256, "threads": 4, "total_ms": 218.39, "total_ms_best": 218.27, "total_ms_worst": 218.57, "total_tps": 1465.283, "total_tps_best": 1466.074, "total_tps_worst": 1464.078, "ttft_ms": 34.62, "warmup": 64}, "prefill_tps": [7388.593858231355, 7399.271634198509, 7399.057776236308], "seq_len": 256}, {"decode_tps": [341.24597436389615, 341.4589902417423, 341.5264095968921], "native": {"agg": "mean", "decode_ms": 187.46, "decode_ms_best": 187.39, "decode_ms_worst": 187.55, "decode_n": 64, "decode_tps": 341.41, "prefill_ms": 82.72, "prefill_ms_best": 82.56, "prefill_ms_worst": 82.83, "prefill_tps": 6189.53, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.98, "samples": {"decode_ms": [187.548, 187.431, 187.394], "prefill_ms": [82.556, 82.778, 82.827], "total_ms": [270.104, 270.209, 270.221], "ttft_ms": [82.556, 82.778, 82.827]}, "seq_len": 512, "threads": 4, "total_ms": 270.18, "total_ms_best": 270.1, "total_ms_worst": 270.22, "total_tps": 2131.927, "total_tps_best": 2132.513, "total_tps_worst": 2131.587, "ttft_ms": 82.72, "warmup": 64}, "prefill_tps": [6201.850864867484, 6185.218294715987, 6181.559153416157], "seq_len": 512}, {"decode_tps": [324.3528147742707, 324.14089999746767, 324.15731759819687], "native": {"agg": "mean", "decode_ms": 197.4, "decode_ms_best": 197.32, "decode_ms_worst": 197.45, "decode_n": 64, "decode_tps": 324.217, "prefill_ms": 218.86, "prefill_ms_best": 218.82, "prefill_ms_worst": 218.94, "prefill_tps": 4678.759, "repeats": 3, "repeats_requested": 3, "rss_mb": 289.98, "samples": {"decode_ms": [197.316, 197.445, 197.435], "prefill_ms": [218.821, 218.938, 218.825], "total_ms": [416.137, 416.384, 416.26], "ttft_ms": [218.822, 218.938, 218.825]}, "seq_len": 1024, "threads": 4, "total_ms": 416.26, "total_ms_best": 416.14, "total_ms_worst": 416.38, "total_tps": 2613.749, "total_tps_best": 2614.522, "total_tps_worst": 2612.976, "ttft_ms": 218.86, "warmup": 64}, "prefill_tps": [4679.623984900901, 4677.123203829395, 4679.53844396207], "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": 0.0911753121984072, "mad_tps": 0.2958144200616175, "median_tps": 324.44574406051254, "samples": 12, "vs_llama_pct": -7.5439790776468225}, "prefill": {"mad_pct": 0.17325623149383484, "mad_tps": 8.131416262921903, "median_tps": 4693.289351160366, "samples": 12, "vs_llama_pct": -64.28636603144733}}, "128": {"decode": {"mad_pct": 0.3411881173356524, "mad_tps": 1.192258927560232, "median_tps": 349.4432739541504, "samples": 12, "vs_llama_pct": -9.258493250631217}, "prefill": {"mad_pct": 0.7256967484964225, "mad_tps": 58.448440858916456, "median_tps": 8054.113647334964, "samples": 12, "vs_llama_pct": -3.484079144155716}}, "256": {"decode": {"mad_pct": 0.19378820892504703, "mad_tps": 0.675455653364736, "median_tps": 348.5535353835626, "samples": 12, "vs_llama_pct": -8.109741142964932}, "prefill": {"mad_pct": 0.0707917821744097, "mad_tps": 5.234068193021358, "median_tps": 7393.609868623147, "samples": 12, "vs_llama_pct": -35.07688518896751}}, "512": {"decode": {"mad_pct": 0.148084595177351, "mad_tps": 0.5064480537859595, "median_tps": 341.99914797310316, "samples": 12, "vs_llama_pct": -7.30285939594838}, "prefill": {"mad_pct": 0.07274005921847713, "mad_tps": 4.5119679336198715, "median_tps": 6202.865356581618, "samples": 12, "vs_llama_pct": -55.927247069402995}}}, "llama.cpp": {"1024": {"decode": {"mad_pct": 0.804031699622986, "mad_tps": 2.821499999999986, "median_tps": 350.919, "samples": 12}, "prefill": {"mad_pct": 0.30399993912390877, "mad_tps": 39.94999999999891, "median_tps": 13141.45, "samples": 12}}, "128": {"decode": {"mad_pct": 1.0377891313238923, "mad_tps": 3.996500000000026, "median_tps": 385.09749999999997, "samples": 12}, "prefill": {"mad_pct": 16.51005320044507, "mad_tps": 1377.7400000000007, "median_tps": 8344.855, "samples": 12}}, "256": {"decode": {"mad_pct": 0.7169502919736922, "mad_tps": 2.7195000000000107, "median_tps": 379.315, "samples": 12}, "prefill": {"mad_pct": 16.236157339293865, "mad_tps": 1849.0150000000003, "median_tps": 11388.255, "samples": 12}}, "512": {"decode": {"mad_pct": 0.3745841154109317, "mad_tps": 1.3819999999999766, "median_tps": 368.9425, "samples": 12}, "prefill": {"mad_pct": 0.2845642543244188, "mad_tps": 40.05000000000018, "median_tps": 14074.15, "samples": 12}}}}}