{ "baseline": { "text_requests": 8, "eos_requests": 8, "strict_json": { "4": false, "5": true, "6": true, "7": true }, "tool_json_passes": 2, "code": { "2": { "pass": true, "error": null }, "3": { "pass": true, "error": null } }, "tool_api_passes": 2, "add_execution_chain_pass": true, "runtime_nll": 2.2528373012744325, "heldout_tokens": 3845, "vision_output": "red", "vision_red_pass": true }, "nvfp4": { "text_requests": 8, "eos_requests": 8, "strict_json": { "4": false, "5": true, "6": true, "7": true }, "tool_json_passes": 2, "code": { "2": { "pass": true, "error": null }, "3": { "pass": true, "error": null } }, "tool_api_passes": 2, "add_execution_chain_pass": true, "runtime_nll": 2.2620555798396893, "heldout_tokens": 3845, "vision_output": "red", "vision_red_pass": true }, "checks": { "export_integrity": true, "all_text_requests_end_normally": true, "both_tool_json_cases_pass": true, "both_tool_api_cases_pass": true, "add_execution_chain_pass": true, "both_code_cases_pass": true, "strict_json_not_below_baseline": true, "heldout_nll_within_15_percent": true, "synthetic_red_image_pass": true }, "relative_runtime_nll_change": 0.004091852776071381, "perplexity_ratio": 1.0092608977529622, "limits": "16 short held-out news articles; 8 text fixtures; one synthetic image. Not a general model, tool-execution, vision, audio, video, long-context, or throughput benchmark." }