This model repo contains the PEFT LoRA adapter from the selected 128-episode Qwen3-Omni diagnostic run in the Ropedia Xperience-10M Task Suite by Chaoyue He.

It is a reproducible baseline and error-analysis artifact. It is not a production robot policy, not a final embodied foundation model, and not a standalone copy of the Qwen3-Omni base model.

Role In The Project

Project layer: Line 2, the selected-128 public-safe comparison surface.
Method row: current Qwen3-Omni v6 LoRA diagnostic row in the 20-task matrix.
Purpose: test structured multimodal JSON output, episode-level split discipline, public-safe packaging, and model-family failure modes.
Comparison context: metadata/raw baselines, Cosmos3 diagnostics, and the single-episode task-head baselines.

Run Identity

Target repo: cy0307/ropedia-qwen3-omni-lora-128ep
Dataset run: xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora
Train run: xperience10m_qwen3_omni_128ep_multiscale_cap96_v6_rank64_lr5e5_full8gpu_lora
Eval run: xperience10m_qwen3_omni_128ep_multiscale_cap96_v6_rank64_lr5e5_full8gpu_lora_eval_test_full
Dataset contract: xperience10m_episode_json_qa_v1
Objective: structured_episode_understanding_json_qa

Adapter And Data Scope

Base model: Qwen/Qwen3-Omni-30B-A3B-Instruct
Adapter method: LoRA, rank 64, alpha 128, dropout 0.05, bf16.
Train split: 96 selected episodes, 25,629 exported windows.
Validation split: 16 selected episodes, 4,608 exported windows.
Test split: 16 selected episodes, 4,032 exported windows.
Raw Xperience-10M MP4/HDF5/RRD files and Qwen3 base weights are not included.

Held-Out Test Metrics

Metric	Value
JSON validity	0.9990
Action macro-F1	0.0029
Subtask accuracy	0.0037
Transition accuracy	0.9898
Next-action accuracy	0.0431
Contact accuracy	0.8177
Object micro-F1	0.3065
Held-out test episodes	14

Interpretation Note

The high JSON validity shows that the output contract and public-safe evaluation pipeline are working. The weak action/subtask values are intentionally visible: they make this a diagnostic baseline for prompt design, target formatting, task-quality analysis, and next-stage model improvement. Do not cite this card as evidence of final full-corpus Qwen3-Omni model quality.

Related Files

Metrics and public-safe predictions are mirrored in the artifact dataset.
Qwen3-Omni v1-v6 lineage is documented in QWEN3_OMNI_RUN_LINEAGE.md and docs/data/qwen3_omni_run_lineage.json in the GitHub/artifact mirrors.
The 180-record task matrix is in docs/data/task_method_20_result_matrix.json.

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 Qwen3-Omni LoRA diagnostic adapter, cite it as:

bibtex

@misc{he2026ropedia_qwen3_omni_lora_128ep,
  author       = {Chaoyue He},
  title        = {Ropedia Xperience-10M Qwen3-Omni LoRA 128-Episode Diagnostic},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep},
  note         = {Selected-128 Qwen3-Omni v6 LoRA diagnostic adapter 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.