Shar137/two / evaluation /SMOKE_STATICFP8.json
Shar137's picture
download
raw
2.12 kB
{
"pkg": "Coder390-static-fp8-block128-mm-mtp-v1",
"text_only": "Coder390-static-fp8-block128-text-v1",
"checks": {
"sglang_load_pkg_no_patch": true,
"greedy_parity": true,
"sglang_image": true,
"sglang_mtp": true,
"vllm_load_pkg_no_patch_mtp": true,
"vllm_image": true
},
"greedy_parity": [
{
"prompt": "用三句话解释为什么天空是蓝色的。",
"identical": true,
"first_diff_char": null,
"len_pkg": 132,
"len_text": 132,
"textonly_self_identical": true,
"textonly_self_first_diff": null
},
{
"prompt": "Write a Python function that r",
"identical": false,
"first_diff_char": 161,
"len_pkg": 791,
"len_text": 681,
"textonly_self_identical": false,
"textonly_self_first_diff": 161
},
{
"prompt": "一个正方形的周长是 36 厘米,它的面积是多少平方厘米?请写",
"identical": true,
"first_diff_char": null,
"len_pkg": 283,
"len_text": 283,
"textonly_self_identical": false,
"textonly_self_first_diff": 104
},
{
"prompt": "List three differences between",
"identical": true,
"first_diff_char": null,
"len_pkg": 512,
"len_text": 512,
"textonly_self_identical": true,
"textonly_self_first_diff": null
}
],
"image_answer": "\n\n图中有一个红色的正方形和一个蓝色的圆形。",
"mtp_finish": [
"stop",
"stop",
"stop",
"stop"
],
"mtp_accept_len_sglang": 3.375,
"vllm_text": "\n\n正方形的周长是 36 厘米。\n\n1. 先求边长: \n 正方形有 4 条相等的边,所以 \n 边长 = 周长 ÷ 4 = 36 ÷ 4 = 9(厘米)\n\n2. 再求面积: \n 面积 = 边长 × 边长 = 9 × 9 = 81(平方厘米)\n\n答:它的面积是 **81 平方厘米**。",
"vllm_image_answer": "\n\n图中有一个红色的正方形和一个蓝色的圆形。",
"status": "PASS",
"tier": "STATICFP8",
"note": "第一轮(FP8 MTP)接受率为 0 被否决,见 round1-mtpfp8/;本结果为 BF16 MTP 发布包"
}

Xet Storage Details

Size:
2.12 kB
·
Xet hash:
c9e0024595780ad44a9426cbba3a4315783a7a0242125ab1be61cd4675c9cbd6

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.