Skip to content

Inquiry Regarding LoRA Configuration for Fine-Tuning Metric3D with ViT-Giant2 #208

Description

@kwong292521

Great work!!!
I’ve been exploring your Metric3D repository, and I noticed that the ViT-Giant2 model with the RAFTDepthNormalDPTDecoder5 already includes a LoRA module, but it’s currently not activated in the default configuration. I’m planning to fine-tune the model on my own dataset and would like to leverage the LoRA module for parameter-efficient fine-tuning, while still loading the pre-trained model you provided.

Could you please guide me on how to properly configure the model to activate the LoRA module while loading the pre-trained weights? Specifically, what changes should I make to the model configuration to ensure that the LoRA module is activated for fine-tuning?

Additionally, if I decide to fine-tune using the built-in LoRA module, are there any extra considerations or precautions I should keep in mind during finetuning?

Thank you for your time, and I appreciate any insights you can share!

Best regards!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions