Summary of Multi Lora Meets Vision: Merging Multiple Adapters to Create a Multi Task Model, by Ege Kesim et al.
Multi LoRA Meets Vision: Merging multiple adapters to create a multi task model
by Ege Kesim, Selahattin Serdar Helli
First submitted to arxiv on: 21 Nov 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates whether multiple LoRA adapters trained on computer vision tasks can be merged together during inference without compromising performance. The goal is to create multitask models by merging different LoRAs, reducing inference time and avoiding additional retraining. To achieve this, the authors train adapters on six different tasks and evaluate their performance when merged together. For comparison, they use a model with a frozen backbone and finetune its head. Results show that merging adapters can be achieved with minimal loss in performance, even outperforming head finetuning in some cases. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In this paper, researchers explore how to combine multiple LoRA adapters trained on different computer vision tasks during inference. They want to see if they can create a multitask model by combining these adapters without losing accuracy or having to retrain the whole model. To test this idea, they train six separate adapters and then merge them together to see how well they perform. For comparison, they also use a traditional approach where they finetune the head of a pre-trained model. The results show that merging adapters is possible with only a slight drop in performance. |
Keywords
» Artificial intelligence » Inference » Lora
