{ "runtime": "Kapitsa 0.2.31 (84249c0)", "host": "Apple M4 Max, 40-core GPU, 64 GiB", "context_limit": 131072, "sampler": "greedy; thinking disabled", "implementation": "request-level strict JSON Schema via response_format; no weight or tokenizer changes; no response postprocessing", "schema_inputs": "Field types and structural constraints only; expected answers are excluded from requests.", "test_source_sha256": "b0e5013aecfe29e2b966afed82120926b700ffd8860a8129ee08e5cdecb62088", "schema_file": "structured_output_profiles.json", "fixture_seeds": { "original": [ 7319, 173 ], "held_out": [ 27183, 65537, 42017 ] }, "vision": [ { "mode": "dflash2", "lane": "baseline", "passed": 4, "total": 8 }, { "mode": "dflash2", "lane": "schema-r1", "passed": 8, "total": 8 }, { "mode": "dflash2", "lane": "schema-r2", "passed": 8, "total": 8 }, { "mode": "dflash2", "lane": "holdout", "passed": 12, "total": 12 }, { "mode": "mtp3", "lane": "schema-mtp", "passed": 8, "total": 8 } ], "coding_passed": 6, "coding_total": 8, "coding_grader_note": "The same standard-library typing allowance is applied to both models.", "speed": { "text_median_decode_tps": 90.2, "structured_json_median_decode_tps": 128.1, "text_probes": 2, "repeats": 3, "measurement": "Median server-reported decode; code/prose capped at 256 tokens; JSON sequence independently checked for exact correctness. Six text and three JSON measurements per model." } }