The model was presented in the paper
Distilling Neuro-Symbolic Programs into 3D Multi-modal LLMs, which was accepted to ICML 2026. This repository contains the released APEIRIA model checkpoint as a LoRA adapter based on the
Qwen3-VL-8B-Instruct backbone.
APEIRIA introduces a three-stage curriculum to progressively build reasoning capabilities:
By transferring reasoning patterns rather than concept-specific knowledge, APEIRIA preserves transparent reasoning and modular interchangeability of planning and perception components.
Please refer to the
official GitHub repository for environment setup, data preparation, and detailed instructions on running inference or training.
1@inproceedings{mo2026,
2 title={Distilling Neuro-Symbolic Programs into 3D Multi-modal LLMs},
3 author={Mo, Wentao and Liu, Yang},
4 booktitle={International Conference on Machine Learning},
5 year={2026}
6}
This code builds upon previous 3D MLLMs and foundation models, including
Chat-Scene,
SegDINO3D, and
Mask3D. It utilizes the
SGLang library for fast multi-modal generation.
