SEGoS-data / README.md
PKUfudawei's picture
Publish concise SEGoS data card
6e822f9
|
Raw History Blame Contribute Delete
1.56 kB
---
license: apache-2.0
task_categories:
- text-retrieval
tags:
- agents
- skills
- retrieval
- graph
- benchmark
---
# SEGoS-data
Data artifacts for [SE-GoS](https://github.com/PKUfudawei/SEGoS). This dataset
contains benchmark tasks, graph bundles, evolution records, and evaluation
traces. Upstream GoS skill libraries and workspaces are distributed separately.
## Contents
| Data | Location |
|---|---|
| SkillsBench v1.0 tasks | `tasks/` |
| SkillsBench and ALFWorld cold-start graphs | `coldstart/`, `alfworld/coldstart/` |
| Evolved graph bundles | `full87/graphs/`, `heldout/graphs/` |
| Evolution records and deltas | `full87/evolution/`, `heldout/evolution/` |
| Evaluation traces | `full87/traces/`, `heldout/traces/` |
## Download
Download the complete dataset with the [Hugging Face CLI](https://huggingface.co/docs/huggingface_hub/guides/cli):
```bash
hf download PKUfudawei/SEGoS-data --repo-type dataset --local-dir ./SEGoS-data
```
For selective downloads and experiment setup, see the [SE-GoS repository](https://github.com/PKUfudawei/SEGoS).
## Provenance
The task archive is from [SkillsBench v1.0](https://github.com/benchflow-ai/skillsbench). Graphs, evolution records, and traces are produced by the SE-GoS pipeline. The upstream GoS skill libraries and workspaces are available from [graph-of-skills-data](https://huggingface.co/datasets/davidliuk/graph-of-skills-data) and are not mirrored here.
When reproducing an experiment, record the dataset revision and the artifact paths used; assets may change across revisions.