Accio-Lab/occamy-1.0-FP8

1{2  "status": "bounded_checks_completed_with_disclosed_quality_regression",3  "format": "blockwise FP8 E4M3FN 128x128, dynamic activation",4  "runtime": "SGLang 0.5.13.post1",5  "hardware": "H200",6  "text_requests": 8,7  "all_text_normal_stop": true,8  "code_functional_checks": [9    true,10    true11  ],12  "strict_json_checks": [13    false,14    true,15    true,16    true17  ],18  "red_image_correct": true,19  "native_tool_correct_with_qwen3_coder": true,20  "wrong_parser_qwen_observation": "XML tool call remained in content, tool_calls empty; corrected parser retest recorded separately",21  "limitations": [22    "Eight text prompts and one image are smoke checks, not broad quality evaluation.",23    "One strict JSON request includes Markdown fences.",24    "No MTP, Unsloth, or training validated. Only one local add-tool roundtrip was tested; this is not broad agent-workflow validation.",25    "Shared GPU and on-demand compilation: timings are not performance comparisons."26  ],27  "local_add_tool_roundtrip_pass": true,28  "local_add_tool_roundtrip_result": 42,29  "expanded_evaluation": "EXPANDED-VALIDATION.json"30}