| { |
| "$schema": "./benchmark.schema.json", |
| "title": "MODEL SYSTEM EVALUATIONS", |
| "subtitle": "REFERENCE LAYOUT / SYNTHETIC DEMONSTRATION DATA", |
| "higherIsBetter": true, |
| "series": [ |
| {"id": "candidate", "label": "CANDIDATE / MXFP4", "color": "#6f36d9"}, |
| {"id": "control", "label": "BF16 CONTROL", "color": "#171719"}, |
| {"id": "native", "label": "NATIVE RTN", "color": "#707078"}, |
| {"id": "baseline", "label": "BASELINE A", "color": "#aaa7af"}, |
| {"id": "reference", "label": "REFERENCE B", "color": "#d5d2d8"} |
| ], |
| "groups": [ |
| { |
| "name": "STEM", |
| "panels": [ |
| {"name": "MATH EXACT", "values": {"candidate": 62.4, "control": 58.9, "native": 60.1, "baseline": 54.2, "reference": 57.8}}, |
| {"name": "CODE PASS@1", "values": {"candidate": 71.8, "control": 67.3, "native": 69.4, "baseline": 61.5, "reference": 64.9}}, |
| {"name": "SCIENCE QA", "values": {"candidate": 76.2, "control": 73.7, "native": 74.1, "baseline": 69.8, "reference": 72.5}} |
| ] |
| }, |
| { |
| "name": "INSTRUCTION FOLLOWING", |
| "panels": [ |
| {"name": "FORMAT", "values": {"candidate": 84.6, "control": 79.2, "native": 81.4, "baseline": 72.1, "reference": 77.3}}, |
| {"name": "MULTI-TURN", "values": {"candidate": 68.5, "control": 63.2, "native": 65.9, "baseline": 58.4, "reference": 61.7}}, |
| {"name": "CONSTRAINTS", "values": {"candidate": 73.9, "control": 70.6, "native": 71.8, "baseline": 66.2, "reference": 68.5}} |
| ] |
| }, |
| { |
| "name": "TOOL USE", |
| "panels": [ |
| {"name": "FUNCTION CALL", "values": {"candidate": 79.1, "control": 70.8, "native": 75.4, "baseline": 64.2, "reference": 69.1}}, |
| {"name": "SANDBOX", "values": {"candidate": 66.7, "control": 61.1, "native": 63.5, "baseline": 56.9, "reference": 60.2}}, |
| {"name": "RETRIEVAL", "values": {"candidate": 81.3, "control": 76.5, "native": 78.2, "baseline": 71.8, "reference": 75.6}} |
| ] |
| }, |
| { |
| "name": "AGENTIC", |
| "panels": [ |
| {"name": "TASK COMPLETE", "values": {"candidate": 57.8, "control": 49.6, "native": 53.1, "baseline": 45.4, "reference": 50.2}}, |
| {"name": "PATCH ACCEPT", "values": {"candidate": 64.2, "control": 55.8, "native": 59.6, "baseline": 51.7, "reference": 56.1}}, |
| {"name": "BROWSER", "values": {"candidate": 47.9, "control": 39.4, "native": 43.3, "baseline": 35.8, "reference": 41.5}} |
| ] |
| } |
| ] |
| } |
|
|
