Qwen3.5-9B fully fine-tuned for embodied spatial reasoning using SPARC-generated VQA data and an EO-1.5M-augmented mixture.

The training mixture contains SPARC-generated VQA data from ours_adaptive_det_soft_snr_sp8, FSD, RoboPoint, LLaVA-OneVision2, and EO-1.5M. SPARC samples use an annotation-quality threshold of 0.97, are sorted by score, and are capped at 700 samples per object.

The SPARC training data is the filtered release subset, which needs no further SPARC filtering. Its filtering script and release mixture manifest are in irl-kit/SPARC-VQA-Raw. FSD, RoboPoint, LLaVA-OneVision2, and EO-1.5M remain their respective upstream datasets.

Prompt formatting is important for these models. Use the bundled chat_template.jinja through processor.apply_chat_template(..., add_generation_prompt=True), with one user turn containing the image(s) followed by the text question. Disable thinking/reasoning mode to match evaluation.

For a single point, append exactly:

For a trajectory or multiple points, append exactly:

The vision encoder is frozen and the vision projector is trainable. The model was fully fine-tuned for one epoch with a learning rate of 2e-5 and a maximum sequence length of 5600.

This is the strongest evaluated 9B run: local full-benchmark aggregate 0.719. It should be compared separately from SPARC-only mixtures because it includes EO-1.5M.