[ { "name": "mainline64k-27b-q3xl-mtp-np3-q8q8", "model": "C:\\Users\\sjake\\LLM\\models\\unsloth-qwen3.6-27b-mtp-gguf\\Qwen3.6-27B-UD-Q3_K_XL.gguf", "mtp": true, "n_ok": 10, "n_total": 10, "avg_wall_s": 12.799441590000423, "avg_prompt_tps": 571.9423534012651, "avg_predicted_tps": 49.14135616945792, "avg_prompt_ms": 2233.2393, "avg_predicted_ms": 10461.8102, "total_prompt_tokens": 13132, "total_completion_tokens": 5120, "draft_tokens": 4062, "accepted_draft_tokens": 3075, "draft_accept_rate": 0.7570162481536189, "vram_before": { "used_mib": 1151, "free_mib": 23163, "power_w": 29.6, "temp_c": 45 }, "vram_ready": { "used_mib": 18762, "free_mib": 5552, "power_w": 81.06, "temp_c": 47 }, "vram_after": { "used_mib": 19108, "free_mib": 5206, "power_w": 468.12, "temp_c": 68 }, "net_ready_mib": 17611, "props_n_ctx": 21504, "total_slots": 3, "log": "C:\\Users\\sjake\\LLM\\logs\\mainline64k-27b-q3xl-mtp-np3-q8q8-20260526-194600.server.log", "interesting_lines": [ "0.00.577.252 I common_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |", "0.00.577.255 I common_memory_breakdown_print: | - CUDA0 (RTX 3090 Ti) | 24563 = 23285 + (13684 = 13410 + 133 + 140) + -12406 |", "0.00.577.256 I common_memory_breakdown_print: | - Host | 697 = 682 + 0 + 15 |", "0.00.992.077 I common_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |", "0.00.992.080 I common_memory_breakdown_print: | - CUDA0 (RTX 3090 Ti) | 24563 = 21937 + (17026 = 13410 + 3488 + 127) + -14399 |", "0.00.992.081 I common_memory_breakdown_print: | - Host | 697 = 682 + 0 + 15 |", "0.02.454.777 I load_tensors: CPU_Mapped model buffer size = 682.03 MiB", "0.02.454.781 I load_tensors: CUDA0 model buffer size = 13410.41 MiB", "0.07.100.164 I llama_kv_cache: CUDA0 KV buffer size = 2142.00 MiB", "0.07.299.042 I llama_memory_recurrent: CUDA0 RS buffer size = 1346.62 MiB", "0.07.310.725 I sched_reserve: CUDA0 compute buffer size = 127.24 MiB", "0.07.310.729 I sched_reserve: CUDA_Host compute buffer size = 15.89 MiB", "0.07.705.568 I llama_kv_cache: CUDA0 KV buffer size = 133.88 MiB", "0.07.720.020 I sched_reserve: CUDA0 compute buffer size = 140.61 MiB", "0.07.720.024 I sched_reserve: CUDA_Host compute buffer size = 15.89 MiB", "0.09.340.728 I sched_reserve: CUDA0 compute buffer size = 263.75 MiB", "0.09.340.732 I sched_reserve: CUDA_Host compute buffer size = 15.89 MiB", "0.21.241.648 I slot print_timing: id 2 | task 0 | draft acceptance = 0.77114 ( 310 accepted / 402 generated)", "0.31.992.050 I slot print_timing: id 1 | task 205 | draft acceptance = 0.74634 ( 306 accepted / 410 generated)", "0.42.797.355 I slot print_timing: id 0 | task 414 | draft acceptance = 0.75862 ( 308 accepted / 406 generated)", "0.42.928.830 W slot update_slots: id 2 | task 622 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.56.321.669 I slot print_timing: id 2 | task 622 | draft acceptance = 0.78392 ( 312 accepted / 398 generated)", "0.56.458.507 W slot update_slots: id 0 | task 829 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.08.817.866 I slot print_timing: id 0 | task 829 | draft acceptance = 0.76049 ( 308 accepted / 405 generated)", "1.08.952.604 W slot update_slots: id 1 | task 1036 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.21.338.314 I slot print_timing: id 1 | task 1036 | draft acceptance = 0.75430 ( 307 accepted / 407 generated)", "1.21.503.907 W slot update_slots: id 2 | task 1245 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.34.524.225 I slot print_timing: id 2 | task 1245 | draft acceptance = 0.77945 ( 311 accepted / 399 generated)", "1.34.665.119 W slot update_slots: id 0 | task 1451 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.47.410.557 I slot print_timing: id 0 | task 1451 | draft acceptance = 0.80102 ( 314 accepted / 392 generated)", "1.47.545.891 W slot update_slots: id 1 | task 1654 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "2.00.468.533 I slot print_timing: id 1 | task 1654 | draft acceptance = 0.72010 ( 301 accepted / 418 generated)", "2.00.610.882 W slot update_slots: id 2 | task 1869 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "2.16.586.145 I slot print_timing: id 2 | task 1869 | draft acceptance = 0.70118 ( 298 accepted / 425 generated)" ], "errors": [], "jsonl": "C:\\Users\\sjake\\LLM\\logs\\swebench-lite-first10-mainline64k-27b-q3xl-mtp-np3-q8q8-20260526-194600.jsonl" }, { "name": "mainline64k-35a3b-q3xl-mtp-np3-q8q8", "model": "C:\\Users\\sjake\\LLM\\models\\unsloth-qwen3.6-35b-a3b-mtp-gguf\\Qwen3.6-35B-A3B-UD-Q3_K_XL.gguf", "mtp": true, "n_ok": 10, "n_total": 10, "avg_wall_s": 8.052431630000138, "avg_prompt_tps": 703.6676584213677, "avg_predicted_tps": 81.36078570012702, "avg_prompt_ms": 1701.4, "avg_predicted_ms": 6293.9525, "total_prompt_tokens": 13132, "total_completion_tokens": 5120, "draft_tokens": 4159, "accepted_draft_tokens": 3022, "draft_accept_rate": 0.7266169752344314, "vram_before": { "used_mib": 1252, "free_mib": 23062, "power_w": 29.11, "temp_c": 48 }, "vram_ready": { "used_mib": 19037, "free_mib": 5277, "power_w": 83.08, "temp_c": 48 }, "vram_after": { "used_mib": 19198, "free_mib": 5116, "power_w": 131.42, "temp_c": 47 }, "net_ready_mib": 17785, "props_n_ctx": 21504, "total_slots": 3, "log": "C:\\Users\\sjake\\LLM\\logs\\mainline64k-35a3b-q3xl-mtp-np3-q8q8-20260526-194826.server.log", "interesting_lines": [ "0.00.541.862 I common_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |", "0.00.541.867 I common_memory_breakdown_print: | - CUDA0 (RTX 3090 Ti) | 24563 = 23285 + (16095 = 15903 + 66 + 125) + -14817 |", "0.00.541.867 I common_memory_breakdown_print: | - Host | 528 = 515 + 0 + 12 |", "0.00.921.992 I common_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |", "0.00.921.996 I common_memory_breakdown_print: | - CUDA0 (RTX 3090 Ti) | 24563 = 22719 + (17263 = 15903 + 1234 + 125) + -15419 |", "0.00.921.997 I common_memory_breakdown_print: | - Host | 528 = 515 + 0 + 12 |", "0.04.447.608 I load_tensors: CPU_Mapped model buffer size = 515.31 MiB", "0.04.447.609 I load_tensors: CUDA0 model buffer size = 15903.70 MiB", "0.10.129.782 I llama_kv_cache: CUDA0 KV buffer size = 669.38 MiB", "0.10.249.805 I llama_memory_recurrent: CUDA0 RS buffer size = 565.31 MiB", "0.10.259.724 I sched_reserve: CUDA0 compute buffer size = 125.22 MiB", "0.10.259.729 I sched_reserve: CUDA_Host compute buffer size = 12.87 MiB", "0.10.536.756 I llama_kv_cache: CUDA0 KV buffer size = 66.94 MiB", "0.10.548.193 I sched_reserve: CUDA0 compute buffer size = 125.22 MiB", "0.10.548.197 I sched_reserve: CUDA_Host compute buffer size = 12.87 MiB", "0.12.055.874 I sched_reserve: CUDA0 compute buffer size = 248.37 MiB", "0.12.055.879 I sched_reserve: CUDA_Host compute buffer size = 12.87 MiB", "0.18.599.541 I slot print_timing: id 2 | task 0 | draft acceptance = 0.77500 ( 310 accepted / 400 generated)", "0.25.788.914 I slot print_timing: id 1 | task 205 | draft acceptance = 0.75616 ( 307 accepted / 406 generated)", "0.33.372.644 I slot print_timing: id 0 | task 413 | draft acceptance = 0.75000 ( 306 accepted / 408 generated)", "0.33.447.518 W slot update_slots: id 2 | task 623 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.42.571.207 I slot print_timing: id 2 | task 623 | draft acceptance = 0.71599 ( 300 accepted / 419 generated)", "0.42.638.348 W slot update_slots: id 0 | task 842 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.49.584.079 I slot print_timing: id 0 | task 842 | draft acceptance = 0.70853 ( 299 accepted / 422 generated)", "0.49.658.547 W slot update_slots: id 1 | task 1058 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.57.028.700 I slot print_timing: id 1 | task 1058 | draft acceptance = 0.74209 ( 305 accepted / 411 generated)", "0.57.102.471 W slot update_slots: id 2 | task 1269 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.05.302.438 I slot print_timing: id 2 | task 1269 | draft acceptance = 0.74209 ( 305 accepted / 411 generated)", "1.05.376.919 W slot update_slots: id 0 | task 1481 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.13.836.915 I slot print_timing: id 0 | task 1481 | draft acceptance = 0.70283 ( 298 accepted / 424 generated)", "1.13.908.320 W slot update_slots: id 1 | task 1700 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.21.733.672 I slot print_timing: id 1 | task 1700 | draft acceptance = 0.72249 ( 302 accepted / 418 generated)", "1.21.812.837 W slot update_slots: id 2 | task 1914 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.32.068.707 I slot print_timing: id 2 | task 1914 | draft acceptance = 0.65909 ( 290 accepted / 440 generated)" ], "errors": [], "jsonl": "C:\\Users\\sjake\\LLM\\logs\\swebench-lite-first10-mainline64k-35a3b-q3xl-mtp-np3-q8q8-20260526-194826.jsonl" }, { "name": "mainline64k-35a3b-q3xl-nonmtp-np3-q8q8", "model": "C:\\Users\\sjake\\LLM\\models\\unsloth-qwen3.6-35b-a3b-gguf\\Qwen3.6-35B-A3B-UD-Q3_K_XL.gguf", "mtp": false, "n_ok": 10, "n_total": 10, "avg_wall_s": 7.277327670000159, "avg_prompt_tps": 877.7555848446085, "avg_predicted_tps": 89.36095540313923, "avg_prompt_ms": 1397.7045, "avg_predicted_ms": 5826.3127, "total_prompt_tokens": 13132, "total_completion_tokens": 5120, "draft_tokens": 0, "accepted_draft_tokens": 0, "draft_accept_rate": null, "vram_before": { "used_mib": 1248, "free_mib": 23066, "power_w": 29.18, "temp_c": 41 }, "vram_ready": { "used_mib": 18110, "free_mib": 6204, "power_w": 81.94, "temp_c": 43 }, "vram_after": { "used_mib": 18107, "free_mib": 6207, "power_w": 215.13, "temp_c": 56 }, "net_ready_mib": 16862, "props_n_ctx": 21504, "total_slots": 3, "log": "C:\\Users\\sjake\\LLM\\logs\\mainline64k-35a3b-q3xl-nonmtp-np3-q8q8-20260526-195008.server.log", "interesting_lines": [ "0.00.539.592 I common_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |", "0.00.539.596 I common_memory_breakdown_print: | - CUDA0 (RTX 3090 Ti) | 24563 = 23095 + (16522 = 15539 + 857 + 125) + -15053 |", "0.00.539.597 I common_memory_breakdown_print: | - Host | 528 = 515 + 0 + 12 |", "0.03.963.331 I load_tensors: CPU_Mapped model buffer size = 515.31 MiB", "0.03.963.333 I load_tensors: CUDA0 model buffer size = 15539.34 MiB", "0.09.725.542 I llama_kv_cache: CUDA0 KV buffer size = 669.38 MiB", "0.09.771.316 I llama_memory_recurrent: CUDA0 RS buffer size = 188.44 MiB", "0.09.780.799 I sched_reserve: CUDA0 compute buffer size = 125.22 MiB", "0.09.780.804 I sched_reserve: CUDA_Host compute buffer size = 12.87 MiB", "0.31.176.106 W slot update_slots: id 2 | task 1546 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.38.949.396 W slot update_slots: id 0 | task 2065 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.46.144.051 W slot update_slots: id 1 | task 2580 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "0.53.013.538 W slot update_slots: id 2 | task 3096 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.00.779.925 W slot update_slots: id 0 | task 3613 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.07.700.976 W slot update_slots: id 1 | task 4130 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", "1.15.252.573 W slot update_slots: id 2 | task 4646 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)" ], "errors": [], "jsonl": "C:\\Users\\sjake\\LLM\\logs\\swebench-lite-first10-mainline64k-35a3b-q3xl-nonmtp-np3-q8q8-20260526-195008.jsonl" } ]