PEFT methods offer energy-efficient personalization for on-device SLMs

A new research paper evaluates various Parameter-Efficient Fine-Tuning (PEFT) methods for personalizing Small Language Models (SLMs) on consumer GPUs. The study compares five methods—Full Fine-Tuning, LoRA, LoRA+, QLoRA, and BitFit—across different SLM architectures, including Transformer-based models like TinyLlama-1.1B and Qwen3-1.7B, and SSM-based models like Mamba-1.4B and Mamba-2-1.3B. Results indicate that LoRA+ is often the most energy-efficient method, while QLoRA excels in reducing peak VRAM usage for Transformer models, suggesting that optimized PEFT techniques offer a viable path for on-device SLM deployment. AI

IMPACT Optimized PEFT methods like LoRA+ and QLoRA enable more efficient on-device personalization of small language models, reducing VRAM and energy costs.

RANK_REASON Research paper evaluating multiple PEFT methods on various SLMs and benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• BitFit
• Full Fine-Tuning
• LoRA+
• Mamba-1.4B
• Mamba-2-1.3B
• PEFT
• QLoRA
• Qwen3-1.7B
• TinyLlama-1.1B

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →