| # Qwen3-Omni LoRA Pipeline Figure Provenance | |
| This presentation figure was generated with ChatGPT image generation and then | |
| committed as a public visual asset for the Ropedia Xperience-10M task-suite | |
| project. | |
| Figure file: | |
| - `docs/assets/qwen3_omni_lora_pipeline.png` | |
| Design brief: | |
| - Show the end-to-end flow from valid Xperience-10M raw episodes into | |
| episode-level split validation, parallel media and sensor export, | |
| Qwen3-Omni-compatible JSONL records, video/audio/text inputs, sensor-bridge | |
| features, LoRA adapter training, and final outputs. | |
| - Make clear that train/validation staging and held-out test evaluation are | |
| separate steps. | |
| - Use the dark green Ropedia-style visual language already used by the website. | |
| - Do not show raw private data, model weights, or unverifiable metrics. | |
| Scope: | |
| - The figure documents the prepared Qwen3-Omni LoRA training path. | |
| - It is not a completed held-out training metric report. | |