README.md

---
license: mit
pretty_name: Ropedia Xperience-10M Task Suite Artifacts
tags:
  - embodied-ai
  - multimodal
  - xperience-10m
  - evaluation
  - qwen3-omni
configs:
  - config_name: episode_sample
    data_files:
      - split: public_sample
        path: viewer/episode_windows.jsonl
---

# Ropedia Xperience-10M Task Suite Artifacts

![Ropedia Xperience-10M Task Suite logo](assets/brand/xperience10m-logo-social-card.png)

This dataset repository stores small derived artifacts for the Ropedia
Xperience-10M task-suite project: metrics, predictions, manifests, reports,
figures, website JSON, public-safe Qwen3-Omni diagnostic outputs, and the
Cosmos3-Nano plus Cosmos3-Super diagnostic packages.

## What To Open First

| Reader goal | Artifact |
| --- | --- |
| Understand current scope | `PROJECT_STATUS.md` |
| Navigate all files | `ARTIFACT_GUIDE.md` |
| Interpret task metrics | `RESEARCH_TAKEAWAYS.md` |
| Check evaluation rules | `EVALUATION_PROTOCOL.md` |
| Inspect single-episode task results | `results/episode_task_suite/summary_report.json` |
| Inspect 128-episode same-split baselines | `results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md` |
| Inspect final Qwen3-Omni held-out diagnostic result | `docs/data/omni_finetune_verified_result.json` |
| Compare current versions and model groups | `docs/data/omni_model_comparison.json` |

## 128-Episode Enhancement Pack

The no-new-episode suite push is recorded in `TASK_SUITE_ENHANCEMENT_128.md`
and `docs/data/task_suite_enhancement_128.json`. It recommends
`multiscale_20s10_40s20_80s40`, hierarchical action/subtask targets,
label-normalized scoring, and compact raw-feature shards before adding more
episodes.

## Dataset Boundary

This artifact bundle contains derived artifacts only. It does not redistribute
Raw Xperience-10M videos, raw `annotation.hdf5`, `.rrd` files, private gated
dataset files, full Qwen weights, LoRA adapter weights, or large checkpoints.
Use of the upstream Xperience-10M dataset remains governed by the official
Ropedia/Xperience-10M access terms.

The implemented public-sample task suite uses one public Xperience-10M sample
episode. The selected 128-episode Qwen3-Omni final diagnostic result uses a
gated local dataset copy and publishes only public-safe metrics, predictions,
manifests, reports, and audits. The Qwen3-Omni LoRA adapter weights are
published separately at `cy0307/ropedia-qwen3-omni-lora-128ep`.
Cosmos3-Nano remains an artifacts-only compatibility result. Cosmos3-Super
Forward-Dynamics LoRA has a separate weight-bearing model repo at
`cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep`.

Source alignment is mirrored through `xperience10m_dataset_card_alignment.json`
and `source_alignment_audit.json`. The official gated dataset card records
`31.9 TB` on the live HF surface and an `about-1PB` full-scale storage
statement; the committed API-listing snapshot records `12,103 episode folders`
as upstream `metadata only`, not local raw-data possession. The public sample
remains under `cc-by-nc-4.0`, with the `HOMIE Toolkit` and `Rerun 0.29.0`
noted as source tooling. The official note that the data is `limited in diversity`
is preserved.

## Derived Artifacts

This bundle includes derived artifacts such as:

- 12-task single-episode metrics, predictions, feature manifests, and neural
  MLP result directories.
- Audio-ablation summaries and generated chart assets.
- Public website JSON and figure manifests.
- A current verified Qwen3-Omni strict-label v3 held-out package for the
  selected 96/16/16 episode split, with 100.00% JSON validity and weak
  action/subtask quality documented as the next error-analysis target.
- Historical Qwen3-Omni packages, including the earlier v2 strict-JSON
  diagnostic, for regression and prompt-contract comparison.
- Verified Cosmos3-Nano future-window compatibility, Cosmos3-Super
  base-weight Reasoner evaluation, and Cosmos3-Super Forward-Dynamics LoRA
  public-safe packages for the same selected split family.
- 128-episode same-split simple/NN metadata baselines for the same 12 task ids,
  with unsupported markers where raw 128 sensor feature blocks are still needed.
- A model-family grouped comparison that pairs 1-episode and 128-episode entries
  for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super without
  mixing target types.

## Related Hub Repositories

| Surface | URL |
| --- | --- |
| HF Space | https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite |
| Artifact dataset | https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts |
| Baseline model repo | https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines |
| Qwen3-Omni LoRA adapter repo | https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep |
| Cosmos3-Super Forward-Dynamics LoRA adapter repo | https://huggingface.co/cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep |
| GitHub repo | https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite |

## Citation

If you use these artifacts, cite this project and the upstream Ropedia
Xperience-10M dataset according to the citation guidance in `CITATION.cff` and
the official dataset card.