Ropedia Xperience-10M Weights, Results, and Analysis

This repo is the consolidated public-safe model-artifact and result package for the Ropedia Xperience-10M Task Suite by Chaoyue He. It makes the current baseline, Qwen3-Omni, Cosmos3-Super, and analysis artifacts easy to find from one Hugging Face page while preserving the canonical per-adapter repos.

What This Repo Contains

• baselines_and_analysis_snapshot/: the baseline-model snapshot with minimal numpy weights, neural MLP checkpoints, metrics, result JSON, and analysis cards from cy0307/ropedia-xperience-10m-task-baselines.
• weights/qwen3_omni_v6_rank64_lora/: current verified Qwen3-Omni v6 rank64 LoRA adapter for the selected-128 diagnostic row.
• weights/qwen3_omni_legacy_lora/: earlier Qwen3-Omni adapter copy retained for provenance.
• weights/cosmos3_super_forward_dynamics_lora/: Cosmos3-Super forward-dynamics LoRA adapter over camera-pose proxy targets.
• analysis/: compact analysis files, Qwen3-Omni v1-v6 lineage, result matrix, and public-safe comparison records.
• results/: verified result summaries, package audits, metrics, and run reports corresponding to the included adapters.
• manifest.json: file-level size and SHA-256 manifest for the staged bundle.

Current Result Coverage

• Method-task records: 180/180 scored.
• Line 1: 1 public sample episode, 2 methods x 20 tasks = 40/40 direct scores.
• Line 2: selected 128 episodes, 7 methods x 20 tasks = 140/140 scores.
• Selected-128 scoring split: 134 direct scores + 6 compact-proxy scores.
• Task surface: 20 embodied-AI tasks x 9 method families.

Interpretation Note

This package is for public-safe reproducibility and diagnostics. The Qwen3-Omni and Cosmos rows are model-family diagnostics, not final leaderboard claims. Weak action/subtask results are retained because they identify the next error-analysis and model-improvement stage. Raw metric values are citeable; radar-normalized values are visual summaries only.

Data Boundary

The repo excludes raw Xperience-10M MP4/HDF5/RRD files, gated full-dataset contents, Qwen3 base weights, Cosmos3 base weights, full fine-tune checkpoints, private logs, optimizer state, and generated local cache files.

Related Repositories

• Project website: https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/
• GitHub source: https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite
• 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
• Consolidated weights/results: https://huggingface.co/cy0307/ropedia-xperience-10m-weights-results
• Qwen3-Omni LoRA adapter: https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep
• Cosmos3-Super LoRA adapter: https://huggingface.co/cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep
• Upstream gated Xperience-10M dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m
• Public sample dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample

Citation

If you use this consolidated weights/results package, cite it as:

@misc{he2026ropedia_xperience10m_weights_results,
  author       = {Chaoyue He},
  title        = {Ropedia Xperience-10M Weights, Results, and Analysis},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/cy0307/ropedia-xperience-10m-weights-results},
  note         = {Public-safe weights, results, and analysis package for the Ropedia Xperience-10M Task Suite}
}

Also cite the main Ropedia Xperience-10M Task Suite and the upstream Ropedia Xperience-10M dataset according to their citation guidance.

Reproducibility

The bundle is rebuilt by:

python3 scripts/build_hf_weights_results_bundle.py
python3 scripts/publish_hf_bundles.py --skip-space --skip-artifacts --skip-model

Use manifest.json to verify byte-level integrity after upload.