gwx22/AwareVLN-ck

AwareVLN Checkpoints (GWxuan/AwareVLN)

Model weights for training and evaluation.

Contents

Subdirectory	Description
navila-llama3-8b-8f/	SFT initialization checkpoint (NaVILA-style VILA: Llama-3 8B + SigLIP + mm_projector, 8 frames)
awarevln/	AwareVLN SFT result (trained on R2R/RxR/R2R-follow/RxR-follow/human mixture)

Each folder contains: config.json, llm/, mm_projector/, vision_tower/.

Usage

Download with Hugging Face Hub:

from huggingface_hub import snapshot_download

snapshot_download("gwx22/AwareVLN", local_dir="./ck", allow_patterns=["navila-llama3-8b-8f/*"])
snapshot_download("gwx22/AwareVLN", local_dir="./ck", allow_patterns=["awarevln/*"])