Instructions to use Yiyang-Ian-Li/EgoAlign-3B-Alignment with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Yiyang-Ian-Li/EgoAlign-3B-Alignment with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct") model = PeftModel.from_pretrained(base_model, "Yiyang-Ian-Li/EgoAlign-3B-Alignment") - Notebooks
- Google Colab
- Kaggle
Configuration Parsing Warning:In adapter_config.json: "peft.base_model_name_or_path" must be a string
EgoAlign 3B · egocentric alignment
This repository contains an EgoAlign adapter, not a full model. EgoAlign combines a Qwen2.5-3B-Instruct language backbone with the public Concerto point encoder and uses a numerical agent pose to ground spatial questions in 3D scenes. Built with Qwen and Concerto.
| Item | Value |
|---|---|
| Training stage | egocentric alignment |
| Base construction | EgoAlign code and instructions |
| Adapter license | MIT |
This is the alignment-stage adapter. It is intended for probing the learned egocentric representation or for further fine-tuning, rather than direct SQA3D evaluation.
Use
Build the corresponding 3B base locally with the initializer, then download this adapter:
hf download Yiyang-Ian-Li/EgoAlign-3B-Alignment --local-dir /path/to/EgoAlign-3B-Alignment
The adapter expects EgoAlign's ScanNet point-cloud preprocessing and numerical-pose convention, as described in the code repository. Upstream model weights and datasets are not included here.
- Downloads last month
- 12