Datasets:
Complete evaluation bundle and publish independent release audit
Browse files- README.md +11 -3
- archives/evaluation_code.zip +2 -2
- files_manifest.json +11 -6
- release_info.json +10 -1
- validation/release_audit/summary.json +77 -0
README.md
CHANGED
|
@@ -32,7 +32,7 @@ python EditableDesignBench_result/extract.py
|
|
| 32 |
cd EditableDesignBench_result
|
| 33 |
```
|
| 34 |
|
| 35 |
-
公开下载,无需登录。也可在 Files 页面只下载需要的模型 ZIP;同时下载 `EditableDesignBench_v4_data.zip`,即可取得对应原图、指令和评测标注。
|
| 36 |
|
| 37 |
| 模型 | 文字 | 资产 | ZIP |
|
| 38 |
|---|---:|---:|---|
|
|
@@ -67,7 +67,7 @@ release_info.json # 数据、代码版本和模型配置
|
|
| 67 |
|
| 68 |
## 🧪 使用现有图片做 MLLM 评分
|
| 69 |
|
| 70 |
-
结果包附带本次代码快照 `archives/evaluation_code.zip`,运行 `extract.py` 时会一起解压。无需下载模型权重或重新生成图片。
|
| 71 |
|
| 72 |
下面以 Fibo 为例;六个模型都可使用同一命令,只需替换模型目录。MLLM 允许原图和结果图尺寸不同,两张完整图片都保持原样,不缩放或裁剪:
|
| 73 |
|
|
@@ -92,7 +92,7 @@ python evaluation_code/scripts/mllm_evaluate.py \
|
|
| 92 |
--workers 10 --require-scoring
|
| 93 |
```
|
| 94 |
|
| 95 |
-
输入为完整原图、完整结果图和自然语言指令,不提供 mask 或目标图给评委。每例一次请求返回四维评分;重复命令会复用已成功的评分,只补失败/缺失项。上述命令会实际调用使用者配置的评委 API;发布的模型图片不需要重新生成。
|
| 96 |
|
| 97 |
### JoyAI:MLLM 与像素指标的尺寸规则不同
|
| 98 |
|
|
@@ -129,6 +129,14 @@ python evaluation_code/scripts/mllm_evaluate.py \
|
|
| 129 |
|
| 130 |
含重试共 13 次 API 调用。这里只验证评分流程,不代表模型全量表现。逐例分数及日志摘要见 [检查记录](validation/other_models_gemini10/README.md)。
|
| 131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
## 📊 指标与版本
|
| 133 |
|
| 134 |
- 文字和资产分开统计;四项像素指标均覆盖各模型全部 200 条任务。
|
|
|
|
| 32 |
cd EditableDesignBench_result
|
| 33 |
```
|
| 34 |
|
| 35 |
+
公开下载,无需登录。也可在 Files 页面只下载需要的模型 ZIP;同时下载 `EditableDesignBench_v4_data.zip`,即可取得对应原图、指令和评测标注。 单独下载时,运行评测还需下载 `evaluation_code.zip`。
|
| 36 |
|
| 37 |
| 模型 | 文字 | 资产 | ZIP |
|
| 38 |
|---|---:|---:|---|
|
|
|
|
| 67 |
|
| 68 |
## 🧪 使用现有图片做 MLLM 评分
|
| 69 |
|
| 70 |
+
结果包附带本次代码快照 `archives/evaluation_code.zip`,运行 `extract.py` 时会一起解压。无需下载模型权重或重新生成图片。 已在 Python 3.12 环境验证,纯 API 评分不需要显卡。
|
| 71 |
|
| 72 |
下面以 Fibo 为例;六个模型都可使用同一命令,只需替换模型目录。MLLM 允许原图和结果图尺寸不同,两张完整图片都保持原样,不缩放或裁剪:
|
| 73 |
|
|
|
|
| 92 |
--workers 10 --require-scoring
|
| 93 |
```
|
| 94 |
|
| 95 |
+
输入为完整原图、完整结果图和自然语言指令,不提供 mask 或目标图给评委。每例一次请求返回四维评分;重复命令会复用已成功的评分,只补失败/缺失项。上述命令会实际调用使用者配置的评委 API;发布的模型图片不需要重新生成。 遇到接口限流或超时时,可重跑同一命令并将 `--workers 10` 改为 `--workers 2`,已成功的维度不会重复请求。
|
| 96 |
|
| 97 |
### JoyAI:MLLM 与像素指标的尺寸规则不同
|
| 98 |
|
|
|
|
| 129 |
|
| 130 |
含重试共 13 次 API 调用。这里只验证评分流程,不代表模型全量表现。逐例分数及日志摘要见 [检查记录](validation/other_models_gemini10/README.md)。
|
| 131 |
|
| 132 |
+
### ✅ 发布包复查
|
| 133 |
+
|
| 134 |
+
已从公开仓库重新下载代码包,在独立目录运行 **279 项离线测试,全部通过**;并检查全部 **1,200 组**原图、任务指令、预测图与原始输出的完整性、对应关系及尺寸。检查覆盖缓存恢复、失败重试、联合评分和不同尺寸图片输入。
|
| 135 |
+
|
| 136 |
+
六个模型共有 26 条真实 Gemini API 实测全部成功;这不等同于已完成 1,200 条全量 MLLM 评分。依赖安装在既有 Python 3.12 环境验证,未测试每一种新机器环境;接口连接、配额等外部故障仍按重试与断点续跑处理。像素指标与 MLLM 的尺寸规则仍分别遵循上文说明。
|
| 137 |
+
|
| 138 |
+
本轮补齐了代码 ZIP 中遗漏的公开配置示例和可选依赖清单;已下载旧版的用户更新 `archives/evaluation_code.zip` 即可。完整记录见 [发布包检查](validation/release_audit/summary.json)。
|
| 139 |
+
|
| 140 |
## 📊 指标与版本
|
| 141 |
|
| 142 |
- 文字和资产分开统计;四项像素指标均覆盖各模型全部 200 条任务。
|
archives/evaluation_code.zip
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:859d846effad3bd1eb6f8fa0646d3ceac0570dac3962cacf5538a235d0b1dde5
|
| 3 |
+
size 256672
|
files_manifest.json
CHANGED
|
@@ -12,8 +12,8 @@
|
|
| 12 |
},
|
| 13 |
{
|
| 14 |
"path": "README.md",
|
| 15 |
-
"bytes":
|
| 16 |
-
"sha256": "
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"path": "archives/EditableDesignBench_v4_data.zip",
|
|
@@ -22,8 +22,8 @@
|
|
| 22 |
},
|
| 23 |
{
|
| 24 |
"path": "archives/evaluation_code.zip",
|
| 25 |
-
"bytes":
|
| 26 |
-
"sha256": "
|
| 27 |
},
|
| 28 |
{
|
| 29 |
"path": "archives/fibo2-edit-turbo_run3.zip",
|
|
@@ -87,8 +87,8 @@
|
|
| 87 |
},
|
| 88 |
{
|
| 89 |
"path": "release_info.json",
|
| 90 |
-
"bytes":
|
| 91 |
-
"sha256": "
|
| 92 |
},
|
| 93 |
{
|
| 94 |
"path": "validation/joyai_gemini16/README.md",
|
|
@@ -129,6 +129,11 @@
|
|
| 129 |
"path": "validation/other_models_gemini10/summary.json",
|
| 130 |
"bytes": 8781,
|
| 131 |
"sha256": "58e3c6c8f4c4f20955e2ce85cd05b7a99a16ce337f81553ac1018fb0af60e37e"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
}
|
| 133 |
],
|
| 134 |
"images": 1200,
|
|
|
|
| 12 |
},
|
| 13 |
{
|
| 14 |
"path": "README.md",
|
| 15 |
+
"bytes": 9009,
|
| 16 |
+
"sha256": "c8f2039de4f3840369e13470251774a8c171a0501ffaa7845e0048bf0b5168f8"
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"path": "archives/EditableDesignBench_v4_data.zip",
|
|
|
|
| 22 |
},
|
| 23 |
{
|
| 24 |
"path": "archives/evaluation_code.zip",
|
| 25 |
+
"bytes": 256672,
|
| 26 |
+
"sha256": "859d846effad3bd1eb6f8fa0646d3ceac0570dac3962cacf5538a235d0b1dde5"
|
| 27 |
},
|
| 28 |
{
|
| 29 |
"path": "archives/fibo2-edit-turbo_run3.zip",
|
|
|
|
| 87 |
},
|
| 88 |
{
|
| 89 |
"path": "release_info.json",
|
| 90 |
+
"bytes": 3184,
|
| 91 |
+
"sha256": "30cea50acbeb8541cef7406032418bc62c04edca6a262b0e8c9f68c90b6bffe0"
|
| 92 |
},
|
| 93 |
{
|
| 94 |
"path": "validation/joyai_gemini16/README.md",
|
|
|
|
| 129 |
"path": "validation/other_models_gemini10/summary.json",
|
| 130 |
"bytes": 8781,
|
| 131 |
"sha256": "58e3c6c8f4c4f20955e2ce85cd05b7a99a16ce337f81553ac1018fb0af60e37e"
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"path": "validation/release_audit/summary.json",
|
| 135 |
+
"bytes": 2432,
|
| 136 |
+
"sha256": "1173c4555c9e6a9e367dbe958517b4be6c4f253cf3be32254d295900fe2903b2"
|
| 137 |
}
|
| 138 |
],
|
| 139 |
"images": 1200,
|
release_info.json
CHANGED
|
@@ -62,5 +62,14 @@
|
|
| 62 |
"cases_per_model": 2,
|
| 63 |
"complete": true,
|
| 64 |
"full_benchmark": false
|
| 65 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
}
|
|
|
|
| 62 |
"cases_per_model": 2,
|
| 63 |
"complete": true,
|
| 64 |
"full_benchmark": false
|
| 65 |
+
},
|
| 66 |
+
"evaluation_bundle_revision": 2,
|
| 67 |
+
"release_audit": {
|
| 68 |
+
"path": "validation/release_audit/summary.json",
|
| 69 |
+
"offline_tests": 279,
|
| 70 |
+
"input_pairs_verified": 1200,
|
| 71 |
+
"real_api_cases": 26,
|
| 72 |
+
"fresh_environment_install_tested": false
|
| 73 |
+
},
|
| 74 |
+
"evaluation_bundle_note": "Packaging completeness fix adds public config examples and optional dependency manifests; inference/scoring code and model output archives are unchanged."
|
| 75 |
}
|
validation/release_audit/summary.json
ADDED
|
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"status": "passed",
|
| 3 |
+
"public_repo": "yomi-chen/EditableDesignBench_result",
|
| 4 |
+
"revision": "0cc8d785eb74d758d2358c4997a41a69c275fd4c",
|
| 5 |
+
"anonymous_download": true,
|
| 6 |
+
"code_downloaded_and_relocated": true,
|
| 7 |
+
"offline_tests_exit_code": 0,
|
| 8 |
+
"offline_test_count": 279,
|
| 9 |
+
"models": [
|
| 10 |
+
{
|
| 11 |
+
"model": "fibo2-edit-turbo",
|
| 12 |
+
"cases": 200,
|
| 13 |
+
"raw_outputs": 200,
|
| 14 |
+
"predictions": 200,
|
| 15 |
+
"all_valid": true
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"model": "flux3-klein",
|
| 19 |
+
"cases": 200,
|
| 20 |
+
"raw_outputs": 200,
|
| 21 |
+
"predictions": 200,
|
| 22 |
+
"all_valid": true
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"model": "joyai-image-edit-plus",
|
| 26 |
+
"cases": 200,
|
| 27 |
+
"raw_outputs": 200,
|
| 28 |
+
"predictions": 200,
|
| 29 |
+
"all_valid": true
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"model": "qwen-image21",
|
| 33 |
+
"cases": 200,
|
| 34 |
+
"raw_outputs": 200,
|
| 35 |
+
"predictions": 200,
|
| 36 |
+
"all_valid": true
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"model": "qwen-image21-turbo",
|
| 40 |
+
"cases": 200,
|
| 41 |
+
"raw_outputs": 200,
|
| 42 |
+
"predictions": 200,
|
| 43 |
+
"all_valid": true
|
| 44 |
+
},
|
| 45 |
+
{
|
| 46 |
+
"model": "sensenova-u1.5-8b-mot",
|
| 47 |
+
"cases": 200,
|
| 48 |
+
"raw_outputs": 200,
|
| 49 |
+
"predictions": 200,
|
| 50 |
+
"all_valid": true
|
| 51 |
+
}
|
| 52 |
+
],
|
| 53 |
+
"cases_verified": 1200,
|
| 54 |
+
"public_manifest_file_count": 26,
|
| 55 |
+
"judge_api_calls_this_audit": 0,
|
| 56 |
+
"installed_environment": "Existing editabledesignbench environment; a new dependency installation was not tested.",
|
| 57 |
+
"scope": "Published download/extraction,artifact integrity and all model input coverage; real MLLM checks remain26 cases,not full1200.",
|
| 58 |
+
"pixel_scope": "All1200 published pixel reports exist. JoyAI uses official bucket reference geometry; native pixel CLI does not accept its mismatched dimensions.",
|
| 59 |
+
"completed_at": "2026-11-08T15:25:02.305016+00:00",
|
| 60 |
+
"packaging_issue": "The initial code-only archive omitted config/models.example.json,used by an included custom-model test.",
|
| 61 |
+
"packaging_fix": "Added public config examples and optional generation/lock requirements; no inference/scoring code or images changed.",
|
| 62 |
+
"added_files": [
|
| 63 |
+
"config/models.example.json",
|
| 64 |
+
"config/judge.example.json",
|
| 65 |
+
"requirements-generation.txt",
|
| 66 |
+
"requirements.lock.txt"
|
| 67 |
+
],
|
| 68 |
+
"tested_python": "3.12.16",
|
| 69 |
+
"code_bundle_sha256": "859d846effad3bd1eb6f8fa0646d3ceac0570dac3962cacf5538a235d0b1dde5",
|
| 70 |
+
"real_api_validation": {
|
| 71 |
+
"cases": 26,
|
| 72 |
+
"dimensions": 104,
|
| 73 |
+
"models": 6,
|
| 74 |
+
"all_successful": true,
|
| 75 |
+
"full1200_judge_cases_tested": false
|
| 76 |
+
}
|
| 77 |
+
}
|