{ "provenance": { "kind": "selected fields from saved Thor report; no new model run", "source": "thortest/reports/20261109-051803/results.json", "inspected_on": "2026-11-10", "profile": "quick", "timezone_of_source_timestamps": "not specified" }, "device": { "gpu": "NVIDIA Thor", "compute_cap": "11.0", "driver": "595.78", "cuda_driver_api": "13.2", "l4t": "R39 (release), REVISION: 2.0, GCID: 45755727, BOARD: generic, EABI: aarch64, DATE: Mon Jun 1 09:28:48 PM UTC 2026", "kernel": "6.8.12-1021-tegra", "docker": "29.1.3", "mem_total_gb": 122.9, "mem_available_gb_at_start": 18.3, "disk_free_gb": 505.4 }, "co_resident_model": "Qwen3.8-Flash-Next; available memory at start 18.3 GB", "image": "vllm/vllm-openai@sha256:cca10003856866925653fb4ec45309df385dc15f802dbd47c289eb9c5720f197", "run": { "engine": "vllm", "model": "qwen3-0.6b", "hf_id": "Qwen/Qwen3-0.6B", "status": "partial", "facts": { "weights_gib": 1.12, "kv_cache_tokens": 61952, "kv_cache_gib": 7.79, "max_concurrency_x": 7.56, "torch_compile_s": 19.36, "engine_init_s": 193.18, "backends": [ "Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_AT" ] }, "startup_s": 229.7, "ram_delta_gb": 11.35, "probes": [ { "key": "models", "title": "模型列表 /v1/models", "status": "pass", "detail": "served: ['qwen3-0.6b']", "seconds": 0 }, { "key": "chat_basic", "title": "基础对话（算术）", "status": "pass", "detail": "answer='17 * 23 = 391'", "seconds": 0.17 }, { "key": "chinese", "title": "中文对话", "status": "pass", "detail": "answer='中国的首都是北京。'", "seconds": 0.09 }, { "key": "stream", "title": "流式输出 SSE", "status": "pass", "detail": "35 chunks, ttft=0.053s, finish=stop, usage reported", "seconds": 0.35 }, { "key": "json_schema", "title": "结构化输出 JSON Schema", "status": "fail", "detail": "not valid JSON: '{\\n \"city\": \"Tokyo\",\\n \"country\": \"Japan\",\\n \"population_million\": 37.45555555555555555555555555555555555555555555555555'", "seconds": 1.52 }, { "key": "tool_call", "title": "工具调用 Function Calling", "status": "pass", "detail": "get_weather({\"city\": \"Paris\"}), finish=tool_calls", "seconds": 0.19 }, { "key": "reasoning", "title": "思考模式（reasoning 分离）", "status": "pass", "detail": "reasoning=3077 chars, answer='To determine how long it will take for both pipes A and B to fill the tank toget'", "seconds": 9.01 }, { "key": "long_context", "title": "长上下文大海捞针", "status": "pass", "detail": "prompt_tokens=5769, answer='7429'", "seconds": 5.31 }, { "key": "concurrency", "title": "并发请求", "status": "pass", "detail": "4 parallel requests ok in 0.1s", "seconds": 0.1 } ], "bench": { "calibration": { "tok_per_word": 1.121, "overhead": 49 }, "levels": [ { "scenario": "short", "scenario_title": "短对话", "input_tokens": 512, "output_tokens": 256, "concurrency": 1, "requests": 3, "ok": 3, "errors": [], "wall_s": 5.4, "measured_input_tokens_avg": 513, "output_tok_s": 142.25, "total_tok_s": 427.3, "req_s": 0.556, "ttft_mean_ms": 43.3, "ttft_median_ms": 46.4, "ttft_p99_ms": 46.5, "tpot_median_ms": 6.77, "tpot_p99_ms": 7.13, "itl_p99_ms": 13.82, "per_stream_tok_s": 147.78, "power": { "samples": 11, "ram_peak_gb": 116.4, "gpu_w_avg": 17.52, "gpu_w_peak": 20.13, "vin_w_avg": 58.83, "vin_w_peak": 62.66, "gpu_temp_peak_c": 40.5 }, "tokens_per_joule": 2.418 }, { "scenario": "short", "scenario_title": "短对话", "input_tokens": 512, "output_tokens": 256, "concurrency": 2, "requests": 4, "ok": 4, "errors": [], "wall_s": 3.8, "measured_input_tokens_avg": 514, "output_tok_s": 269.77, "total_tok_s": 810.9, "req_s": 1.054, "ttft_mean_ms": 60.5, "ttft_median_ms": 60.5, "ttft_p99_ms": 63.9, "tpot_median_ms": 7.2, "tpot_p99_ms": 7.21, "itl_p99_ms": 12.11, "per_stream_tok_s": 138.88, "power": { "samples": 8, "ram_peak_gb": 116.4, "gpu_w_avg": 18.58, "gpu_w_peak": 18.95, "vin_w_avg": 55.72, "vin_w_peak": 63.49, "gpu_temp_peak_c": 41.06 }, "tokens_per_joule": 4.842 }, { "scenario": "short", "scenario_title": "短对话", "input_tokens": 512, "output_tokens": 256, "concurrency": 4, "requests": 8, "ok": 8, "errors": [], "wall_s": 4.15, "measured_input_tokens_avg": 511, "output_tok_s": 493.76, "total_tok_s": 1479.58, "req_s": 1.929, "ttft_mean_ms": 89.9, "ttft_median_ms": 89.2, "ttft_p99_ms": 98.8, "tpot_median_ms": 7.77, "tpot_p99_ms": 7.82, "itl_p99_ms": 13.19, "per_stream_tok_s": 128.63, "power": { "samples": 8, "ram_peak_gb": 116.4, "gpu_w_avg": 20.56, "gpu_w_peak": 21.53, "vin_w_avg": 63.86, "vin_w_peak": 66.29, "gpu_temp_peak_c": 41.84 }, "tokens_per_joule": 7.732 }, { "scenario": "long_input", "scenario_title": "长输入", "input_tokens": 4096, "output_tokens": 256, "concurrency": 1, "requests": 3, "ok": 3, "errors": [], "wall_s": 6.87, "measured_input_tokens_avg": 4085, "output_tok_s": 111.85, "total_tok_s": 1896.6, "req_s": 0.437, "ttft_mean_ms": 147.6, "ttft_median_ms": 148.1, "ttft_p99_ms": 148.1, "tpot_median_ms": 8.41, "tpot_p99_ms": 8.41, "itl_p99_ms": 14.42, "per_stream_tok_s": 118.96, "power": { "samples": 14, "ram_peak_gb": 116.4, "gpu_w_avg": 23.7, "gpu_w_peak": 30.33, "vin_w_avg": 63.39, "vin_w_peak": 69.62, "gpu_temp_peak_c": 43.06 }, "tokens_per_joule": 1.764 }, { "scenario": "long_input", "scenario_title": "长输入", "input_tokens": 4096, "output_tokens": 256, "concurrency": 2, "requests": 4, "ok": 4, "errors": [], "wall_s": 5.8, "measured_input_tokens_avg": 4068, "output_tok_s": 176.48, "total_tok_s": 2981.04, "req_s": 0.689, "ttft_mean_ms": 247.4, "ttft_median_ms": 246.7, "ttft_p99_ms": 250, "tpot_median_ms": 10.4, "tpot_p99_ms": 10.41, "itl_p99_ms": 18.15, "per_stream_tok_s": 96.16, "power": { "samples": 11, "ram_peak_gb": 116.4, "gpu_w_avg": 27.24, "gpu_w_peak": 35.07, "vin_w_avg": 73.85, "vin_w_peak": 114.78, "gpu_temp_peak_c": 45.41 }, "tokens_per_joule": 2.39 }, { "scenario": "long_input", "scenario_title": "长输入", "input_tokens": 4096, "output_tokens": 256, "concurrency": 4, "requests": 8, "ok": 8, "errors": [], "wall_s": 8.42, "measured_input_tokens_avg": 4082, "output_tok_s": 243.24, "total_tok_s": 4122.1, "req_s": 0.95, "ttft_mean_ms": 450.2, "ttft_median_ms": 465, "ttft_p99_ms": 704.1, "tpot_median_ms": 14.61, "tpot_p99_ms": 16.02, "itl_p99_ms": 26.42, "per_stream_tok_s": 68.43, "power": { "samples": 17, "ram_peak_gb": 116.59, "gpu_w_avg": 32.5, "gpu_w_peak": 51.71, "vin_w_avg": 81.23, "vin_w_peak": 109.08, "gpu_temp_peak_c": 52.31 }, "tokens_per_joule": 2.995 } ], "prefill": [ { "target_tokens": 2048, "prompt_tokens": 2034, "ttft_s": 0.068, "prefill_tok_s": 29789.8, "error": null }, { "target_tokens": 6144, "prompt_tokens": 6084, "ttft_s": 0.239, "prefill_tok_s": 25441.4, "error": null } ] } }, "incomplete": [ "SGLang image pull was interrupted; no SGLang model results", "Granite download failed; no completed MoE quick run", "standard 8B and 30B-A3B were not measured" ] }