This is a Fine-Tuned LoRA adapter for InternVL2.5-4B. It has been trained on the WalkVLM dataset to assist visually impaired individuals by detecting navigation hazards and providing scene descriptions.

Intended Use

Task: Visual Question Answering for Navigation
Domain: Blind Assistance / Accessibility
Base Model: OpenGVLab/InternVL2_5-4B

Training Details

Framework: PyTorch & InternVL
Method: LoRA (Low-Rank Adaptation)
Dataset: WalkVLM (Custom subset)

How to Use

python

from peft import PeftModel
from transformers import AutoModel, AutoTokenizer

# Load Base Model
model = AutoModel.from_pretrained("OpenGVLab/InternVL2_5-4B", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("OpenGVLab/InternVL2_5-4B", trust_remote_code=True)

# Load this Adapter
model = PeftModel.from_pretrained(model, "blind-assist/internvl2-5-4b-walk-lora-v1-100")

# Now you can use the model for inference!