IBM / terratorch

a Python toolkit for fine-tuning Geospatial Foundation Models (GFMs).
Apache License 2.0
170 stars 24 forks source link

Add LoRA support for fine-tuning #225

Open smandal94 opened 3 weeks ago

smandal94 commented 3 weeks ago

Is your feature request related to a problem? Please describe. With increasing parameter sizes of pre-trained models, it has become necessary to adapt parameter efficient fine-tuning methods like LoRA, to allow people to work in constrained resource environments.

Describe the solution you'd like Add PEFT library support in configure_models() function for different tasks.

romeokienzler commented 3 weeks ago

JJ (Johannes Jakubik) -> put in the loop

romeokienzler commented 3 weeks ago

@smandal94 creating PR (probably WIP)

paolofraccaro commented 2 weeks ago

@blumenstiel and his student have a working prototype

smandal94 commented 2 weeks ago

235