CERN/colliderml-benchmark-results

---
license: mit
tags:
- colliderml
- benchmark
- particle-physics
- tracking
pretty_name: ColliderML Benchmark Results
size_categories:
- n<1K
---

# ColliderML Benchmark Results

Machine-scored leaderboard results for the [ColliderML](https://github.com/OpenDataDetector/ColliderML) benchmark tasks.

## Structure

```
results/
{task}/
{username}/
{predictions_sha256}.json
```

Each JSON file records one scored submission:

| Field | Description |
|---|---|
| `submission_id` | UUID assigned by the backend |
| `task` | Benchmark task name (e.g. `tracking`) |
| `submitter` | HuggingFace username |
| `model_repo_id` | Optional link to the submitter's model repo |
| `submitted_at` | ISO 8601 timestamp |
| `scores` | Dict of metric name → value |
| `credits_earned` | Credits awarded for this submission |
| `is_baseline` | Whether this is a reference baseline |
| `predictions_sha256` | SHA-256 of the submitted parquet |

## Tasks

| Task | Primary metric | Dataset |
|---|---|---|
| `tracking` | TrackML weighted efficiency | `ttbar_pu200` |
| `jets` | Jet classification AUC | `ttbar_pu200` |
| `anomaly` | Anomaly AUROC | mixed |
| `tracking_latency` | Inference latency (ms) | `ttbar_pu200` |
| `tracking_small` | TrackML eff under parameter budget | `ttbar_pu200` |
| `data_loading` | Load throughput (events/s) | `ttbar_pu0` |

## Integration

Results are pushed automatically by the ColliderML backend after each scored submission.
The [ColliderML leaderboard Space](https://huggingface.co/spaces/OpenDataDetector/colliderml-leaderboard)
reads from this dataset to render the public table.

Models that include `.eval_results/colliderml_*.yaml` files link back to this dataset,
so scores appear on model cards automatically.

## Links

- **Library:** [OpenDataDetector/ColliderML](https://github.com/OpenDataDetector/ColliderML)
- **Data:** [CERN/ColliderML-Release-1](https://huggingface.co/datasets/CERN/ColliderML-Release-1)
- **Docs:** [opendatadetector.github.io/ColliderML](https://opendatadetector.github.io/ColliderML/)

许可证：MIT
标签：
- ColliderML
- 基准测试
- 粒子物理
- 轨迹追踪
美观名称：ColliderML基准测试结果
样本规模类别：少于1000条
---
# ColliderML基准测试结果

本数据集包含针对[ColliderML](https://github.com/OpenDataDetector/ColliderML)基准测试任务的机器评分排行榜结果。

## 数据集结构

results/
{任务名称}/
{提交者用户名}/
{预测结果SHA256哈希值}.json

每个JSON文件对应一条已评分的提交结果：

| 字段 | 说明 |
|---|---|
| `submission_id` | 后端分配的通用唯一识别码（UUID） |
| `task` | 基准测试任务名称（例如`tracking`） |
| `submitter` | Hugging Face平台用户名 |
| `model_repo_id` | 提交者模型仓库链接（可选） |
| `submitted_at` | ISO 8601格式时间戳 |
| `scores` | 指标名称与对应数值的字典 |
| `credits_earned` | 本次提交获得的积分 |
| `is_baseline` | 是否为参考基线模型 |
| `predictions_sha256` | 提交的Parquet文件的SHA-256哈希值 |

## 基准测试任务

| 任务名称 | 核心指标 | 关联数据集 |
|---|---|---|
| `tracking` | TrackML加权追踪效率 | `ttbar_pu200` |
| `jets` | 喷气粒子分类AUC | `ttbar_pu200` |
| `anomaly` | 异常检测AUROC | 混合数据集 |
| `tracking_latency` | 推理延迟（单位：毫秒） | `ttbar_pu200` |
| `tracking_small` | 参数约束下的TrackML追踪效率 | `ttbar_pu200` |
| `data_loading` | 数据加载吞吐量（事件/秒） | `ttbar_pu0` |

## 数据集集成方式

每次提交结果经评分后，ColliderML后端会自动将结果推送至此数据集。[ColliderML排行榜空间](https://huggingface.co/spaces/OpenDataDetector/colliderml-leaderboard)将读取本数据集内容，以渲染公开排行榜表格。

包含`.eval_results/colliderml_*.yaml`文件的模型会自动关联至本数据集，因此其评分将自动展示在模型卡片上。

## 相关链接

- **工具库：** [OpenDataDetector/ColliderML](https://github.com/OpenDataDetector/ColliderML)
- **数据集：** [CERN/ColliderML-Release-1](https://huggingface.co/datasets/CERN/ColliderML-Release-1)
- **官方文档：** [opendatadetector.github.io/ColliderML](https://opendatadetector.github.io/ColliderML/)