LayerStream — Edge Out-of-Core LLM Benchmarks
Collection
Bare-metal Orin Nano benchmarks, interactive Space, and tested models for layerstream-plugin (zero-patch llama.cpp NVMe & MoE streaming). • 8 items • Updated
record_id stringlengths 25 40 | timestamp_utc stringlengths 32 32 | model_slug stringclasses 2
values | model_name stringclasses 2
values | hf_source stringclasses 2
values | hf_target_repo stringclasses 2
values | params_b float64 4.02 7.62 | quant stringclasses 7
values | imatrix_used bool 2
classes | bpw float64 3.75 8.52 | weight_bytes int64 1.96B 8.1B | weight_gib float64 1.83 7.54 | hardware_id stringclasses 2
values | hardware_label stringclasses 2
values | backend stringclasses 2
values | memory_budget_gib float64 7.42 62 | status stringclasses 2
values | oom_reason stringclasses 3
values | pp512_tok_s float64 4.35 778 ⌀ | pp512_std float64 0.03 17 ⌀ | tg128_tok_s float64 1.05 22.4 ⌀ | tg128_std float64 0.01 0.21 ⌀ | ppl float64 8.27 15.7 | ppl_base_fp16 float64 8.27 14.5 | mean_kld float64 0 0.16 | p99_kld float64 0 2.02 | median_kld float64 0 0.03 | max_kld float64 0.14 13.9 | rms_delta_p_pct float64 0.34 11.1 | same_top1_pct float64 87.5 99.6 | avg_watts float64 18.6 20.9 ⌀ | joules_per_token float64 0.85 1.76 ⌀ | peak_mem_mib float64 2.05k 7.9k ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
qwen2.5-coder-7b__nixos-nvme__Q8_0 | 2026-11-14T17:00:00.749854+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q8_0 | true | 8.5 | 8,098,526,208 | 7.542 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 5.82 | 0.12 | 2.2 | 0.04 | 8.275998 | 8.2694 | 0.000258 | 0.001057 | 0.000053 | 0.195246 | 0.341 | 99.608 | null | null | 7,903.4 |
qwen2.5-coder-7b__nixos-nvme__Q6_K | 2026-11-14T17:08:22.836975+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q6_K | true | 6.57 | 6,254,199,808 | 5.825 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 6.03 | 0.03 | 2.6 | 0.08 | 8.277826 | 8.2694 | 0.001212 | 0.01087 | 0.000442 | 0.139522 | 0.903 | 98.693 | null | null | 6,144.5 |
qwen2.5-coder-7b__nixos-nvme__Q5_K_M | 2026-11-14T17:17:50.594210+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q5_K_M | true | 5.72 | 5,444,832,256 | 5.071 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 5.41 | 0.3 | 2.05 | 0.15 | 8.267634 | 8.2694 | 0.002858 | 0.024508 | 0.00104 | 0.27066 | 1.472 | 97.974 | null | null | 5,372.6 |
qwen2.5-coder-7b__nixos-nvme__Q4_K_M-std | 2026-11-14T17:32:14.873920+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q4_K_M-std | false | 4.92 | 4,683,074,240 | 4.361 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 7.4 | 0.49 | 3.01 | 0.1 | 8.281188 | 8.2694 | 0.013208 | 0.100173 | 0.006151 | 0.226645 | 3.043 | 95.49 | null | null | 4,646.1 |
qwen2.5-coder-7b__nixos-nvme__Q4_K_M | 2026-11-14T17:25:16.665944+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q4_K_M | true | 4.92 | 4,683,074,560 | 4.361 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 7.03 | 0.67 | 2.75 | 0.21 | 8.355934 | 8.2694 | 0.010053 | 0.084548 | 0.004183 | 0.444835 | 2.887 | 96.078 | null | null | 4,646.1 |
qwen2.5-coder-7b__nixos-nvme__IQ4_XS | 2026-11-14T17:41:11.704030+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | IQ4_XS | true | 4.43 | 4,218,473,472 | 3.929 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 5.53 | 0.69 | 2.35 | 0.19 | 8.307361 | 8.2694 | 0.012872 | 0.10344 | 0.005416 | 0.551553 | 3.067 | 95.359 | null | null | 4,203 |
qwen2.5-coder-7b__nixos-nvme__IQ3_M | 2026-11-14T17:55:09.395746+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | IQ3_M | true | 3.75 | 3,574,012,928 | 3.329 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 4.35 | 0.24 | 1.05 | 0.07 | 8.373718 | 8.2694 | 0.041866 | 0.292807 | 0.016326 | 3.880723 | 6.497 | 90.523 | null | null | 3,588.4 |
qwen2.5-coder-7b__orin-nano__Q8_0 | 2026-11-14T18:44:14.334485+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q8_0 | true | 8.5 | 8,098,526,208 | 7.542 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | oom | Weights (7.54 GiB) + CUDA/KV context exceed 5.05 GiB available unified RAM | null | null | null | null | 8.275998 | 8.2694 | 0.000258 | 0.001057 | 0.000053 | 0.195246 | 0.341 | 99.608 | null | null | null |
qwen2.5-coder-7b__orin-nano__Q6_K | 2026-11-14T18:44:14.335434+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q6_K | true | 6.57 | 6,254,199,808 | 5.825 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | oom | Weights (5.83 GiB) + CUDA/KV context exceed 5.05 GiB available unified RAM | null | null | null | null | 8.277826 | 8.2694 | 0.001212 | 0.01087 | 0.000442 | 0.139522 | 0.903 | 98.693 | null | null | null |
qwen2.5-coder-7b__orin-nano__Q5_K_M | 2026-11-14T18:44:14.336596+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q5_K_M | true | 5.72 | 5,444,832,256 | 5.071 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | oom | Weights (5.07 GiB) + CUDA/KV context exceed 5.05 GiB available unified RAM | null | null | null | null | 8.267634 | 8.2694 | 0.002858 | 0.024508 | 0.00104 | 0.27066 | 1.472 | 97.974 | null | null | null |
qwen2.5-coder-7b__orin-nano__Q4_K_M-std | 2026-11-14T18:51:16.923783+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q4_K_M-std | false | 4.92 | 4,683,074,240 | 4.361 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 437 | 3.35 | 12.96 | 0.01 | 8.281188 | 8.2694 | 0.013208 | 0.100173 | 0.006151 | 0.226645 | 3.043 | 95.49 | 20.29 | 1.565 | 5,307.2 |
qwen2.5-coder-7b__orin-nano__Q4_K_M | 2026-11-14T18:47:27.004235+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | Q4_K_M | true | 4.92 | 4,683,074,560 | 4.361 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 437.11 | 3.05 | 12.97 | 0.01 | 8.355934 | 8.2694 | 0.010053 | 0.084548 | 0.004183 | 0.444835 | 2.887 | 96.078 | 20.36 | 1.5701 | 5,017.9 |
qwen2.5-coder-7b__orin-nano__IQ4_XS | 2026-11-14T18:54:31.369657+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | IQ4_XS | true | 4.43 | 4,218,473,472 | 3.929 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 514.91 | 5.5 | 16.3 | 0.02 | 8.307361 | 8.2694 | 0.012872 | 0.10344 | 0.005416 | 0.551553 | 3.067 | 95.359 | 20.47 | 1.2558 | 4,660.3 |
qwen2.5-coder-7b__orin-nano__IQ3_M | 2026-11-14T18:57:29.982638+00:00 | qwen2.5-coder-7b | Qwen2.5-Coder-7B-Instruct | Qwen/Qwen2.5-Coder-7B-Instruct | kleinbem/Qwen2.5-Coder-7B-Instruct-Edge-GGUF | 7.62 | IQ3_M | true | 3.75 | 3,574,012,928 | 3.329 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 437.43 | 2.74 | 11.88 | 0.01 | 8.373718 | 8.2694 | 0.041866 | 0.292807 | 0.016326 | 3.880723 | 6.497 | 90.523 | 20.86 | 1.7564 | 3,828.4 |
qwen3-4b__nixos-nvme__Q8_0 | 2026-11-14T13:28:17.906715+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q8_0 | true | 8.52 | 4,280,405,792 | 3.986 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 9.66 | 0.18 | 3.4 | 0.02 | 14.501104 | 14.4983 | 0.001112 | 0.008848 | 0.000102 | 0.653047 | 1.752 | 98.889 | null | null | 4,262.1 |
qwen3-4b__nixos-nvme__Q6_K | 2026-11-14T13:33:45.978401+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q6_K | true | 6.58 | 3,306,261,792 | 3.079 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 9.06 | 0.12 | 3.93 | 0.1 | 14.325958 | 14.4983 | 0.004937 | 0.063154 | 0.000869 | 0.197927 | 2.277 | 97.647 | null | null | 3,333.1 |
qwen3-4b__nixos-nvme__Q5_K_M | 2026-11-14T13:39:09.063933+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q5_K_M | true | 5.75 | 2,889,514,272 | 2.691 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 9.24 | 0.18 | 3.72 | 0.11 | 14.616385 | 14.4983 | 0.013732 | 0.178855 | 0.002088 | 1.950042 | 4.182 | 96.013 | null | null | 2,935.7 |
qwen3-4b__nixos-nvme__Q4_K_M-std | 2026-11-14T13:47:19.227787+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q4_K_M-std | false | 4.97 | 2,497,280,992 | 2.326 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 12.45 | 0.23 | 4.62 | 0.19 | 15.746229 | 14.4983 | 0.082616 | 0.962711 | 0.010189 | 13.896621 | 7.787 | 91.895 | null | null | 2,561.6 |
qwen3-4b__nixos-nvme__Q4_K_M | 2026-11-14T13:43:10.791345+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q4_K_M | true | 4.97 | 2,497,281,312 | 2.326 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 13.15 | 0.14 | 4.65 | 0.08 | 14.571122 | 14.4983 | 0.04804 | 0.588357 | 0.008513 | 2.603345 | 6.434 | 92.157 | null | null | 2,561.6 |
qwen3-4b__nixos-nvme__IQ4_XS | 2026-11-14T13:52:34.783356+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | IQ4_XS | true | 4.52 | 2,270,752,032 | 2.115 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 9.36 | 0.05 | 3.98 | 0.09 | 15.317879 | 14.4983 | 0.075718 | 0.993617 | 0.009262 | 10.941069 | 9.038 | 92.81 | null | null | 2,345.6 |
qwen3-4b__nixos-nvme__IQ3_M | 2026-11-14T13:58:57.130180+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | IQ3_M | true | 3.91 | 1,962,896,672 | 1.828 | nixos-nvme | Intel Core i3-1315U (Raptor Lake AVX2, 62 GiB) | CPU (AVX2 / AlderLake ggml) | 62 | ok | null | 9.34 | 0.2 | 2.39 | 0.06 | 14.579035 | 14.4983 | 0.155597 | 2.020807 | 0.026007 | 7.501784 | 11.14 | 87.516 | null | null | 2,052 |
qwen3-4b__orin-nano__Q8_0 | 2026-11-14T14:08:37.774571+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q8_0 | true | 8.52 | 4,280,405,792 | 3.986 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 756.26 | 11.14 | 15.2 | 0.01 | 14.501104 | 14.4983 | 0.001112 | 0.008848 | 0.000102 | 0.653047 | 1.752 | 98.889 | 18.6 | 1.2235 | 4,922.1 |
qwen3-4b__orin-nano__Q6_K | 2026-11-14T14:11:47.512285+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q6_K | true | 6.58 | 3,306,261,792 | 3.079 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 598.74 | 7.89 | 14.43 | 0.01 | 14.325958 | 14.4983 | 0.004937 | 0.063154 | 0.000869 | 0.197927 | 2.277 | 97.647 | 19.69 | 1.3646 | 3,631.9 |
qwen3-4b__orin-nano__Q5_K_M | 2026-11-14T14:14:27.514335+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q5_K_M | true | 5.75 | 2,889,514,272 | 2.691 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 657.33 | 7.69 | 16.91 | 0.01 | 14.616385 | 14.4983 | 0.013732 | 0.178855 | 0.002088 | 1.950042 | 4.182 | 96.013 | 19.85 | 1.174 | 3,175.7 |
qwen3-4b__orin-nano__Q4_K_M-std | 2026-11-14T14:18:54.550025+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q4_K_M-std | false | 4.97 | 2,497,280,992 | 2.326 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 676.24 | 9.44 | 18.81 | 0.03 | 15.746229 | 14.4983 | 0.082616 | 0.962711 | 0.010189 | 13.896621 | 7.787 | 91.895 | 19.6 | 1.0422 | 2,801.6 |
qwen3-4b__orin-nano__Q4_K_M | 2026-11-14T14:16:37.508059+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | Q4_K_M | true | 4.97 | 2,497,281,312 | 2.326 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 677.91 | 7.43 | 18.82 | 0.01 | 14.571122 | 14.4983 | 0.04804 | 0.588357 | 0.008513 | 2.603345 | 6.434 | 92.157 | 20.12 | 1.0693 | 2,801.6 |
qwen3-4b__orin-nano__IQ4_XS | 2026-11-14T14:20:59.710509+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | IQ4_XS | true | 4.52 | 2,270,752,032 | 2.115 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 777.6 | 16.97 | 22.39 | 0.01 | 15.317879 | 14.4983 | 0.075718 | 0.993617 | 0.009262 | 10.941069 | 9.038 | 92.81 | 19.1 | 0.8529 | 2,585.6 |
qwen3-4b__orin-nano__IQ3_M | 2026-11-14T14:22:51.353085+00:00 | qwen3-4b | Qwen3-4B | Qwen/Qwen3-4B | kleinbem/Qwen3-4B-Edge-GGUF | 4.02 | IQ3_M | true | 3.91 | 1,962,896,672 | 1.828 | orin-nano | NVIDIA Jetson Orin Nano (Ampere sm_87 CUDA 12.6, 8 GiB Unified) | CUDA 12.6 (Tegra sm_87) | 7.42 | ok | null | 687.29 | 6.74 | 17.83 | 0.01 | 14.579035 | 14.4983 | 0.155597 | 2.020807 | 0.026007 | 7.501784 | 11.14 | 87.516 | 20.41 | 1.1446 | 2,333.2 |
Hardware throughput (pp512, tg128), exact FP16 logit Kullback-Leibler divergence (mean_kld, p99_kld, same_top1_pct), unified-memory OOM boundaries, and INA3221 / RAPL energy per token (Joules/token) collected reproducibly via github.com/kleinbem/hf-bench.
spaces/kleinbem/edge-llm-paretobase_models:
- Qwen/Qwen2.5-Coder-7B-Instruct
- Qwen/Qwen3-4B
orin-nano: NVIDIA Jetson Orin Nano (8 GiB unified LPDDR5, Ampere 1024-core Tegra GPU sm_87, NixOS + JetPack 6 CUDA 12.6, INA3221 hardware power rail sampling).nixos-nvme: 13th Gen Intel Core i3-1315U (x86_64 Raptor Lake AVX2, 62 GiB DDR5, NixOS idle scheduler priority, Intel RAPL package energy sampling).| Column | Type | Description |
|---|---|---|
model_name |
string |
Base model architecture name |
quant |
string |
llama.cpp quantization (Q8_0 .. IQ3_M, plus Q4_K_M-std uncalibrated control) |
imatrix_used |
bool |
Whether agentic + code importance matrix calibration was applied |
bpw |
float64 |
Effective bits per weight (weight_bytes * 8 / params) |
status |
string |
ok or oom when exceeding the device's unified memory cliff |
mean_kld |
float64 |
Mean Kullback-Leibler divergence (nats) vs exact FP16 base logits |
same_top1_pct |
float64 |
Percentage of evaluation tokens sharing the exact FP16 argmax prediction |
pp512_tok_s |
float64 |
512-token prompt prefill speed (tokens/sec) |
tg128_tok_s |
float64 |
128-token autoregressive decode speed (tokens/sec) |
avg_watts |
float64 |
Mean board/package power draw during inference (Watts) |
joules_per_token |
float64 |
Energy consumed per generated token (avg_watts / tg128_tok_s) |