{ "inputs": { "prefill_tps": 6.17, "decode_tps": 4.05, "prompt_tokens": 512, "output_tokens": 256, "tool_overhead_s": 2.0, "turns": 30, "timeout_s": 600.0 }, "outputs": { "prefill_s": 82.98, "decode_s": 63.21, "turn_s": 148.19, "task_s": 4445.8, "task_h": 1.23, "min_decode_tps_for_turn": 0.497, "prompt_tokens_at_timeout": 3702, "turn_fits_timeout": true }, "turn_multipliers": { "prompt_x2_turn_s": 231.17, "prompt_x8_turn_s": 729.07 }, "provenance": { "prefill_tps": "llama-bench README table, llama 7B mostly Q4_0, CPU, 1 thread, pp 64 row", "decode_tps": "llama-bench README table, same model/backend/threads, tg 16 row", "timeout_s": "openai-python DEFAULT_TIMEOUT", "prompt_tokens/output_tokens/tool_overhead_s": "declared scenario of this harness", "caveat": "upstream sample rows, not a measurement of this script or of the reader's host" } }