SLAP: Stratified Loss-based Pruning for On-Policy Data-Efficient Instruction Tuning
Quick Answer
SLAP introduces a batch-aware data selection framework that optimizes instruction tuning for large language models like LLaMA and ChatGLM, achieving 20-40% less training data while maintaining performance.
Quick Take
This method significantly reduces computational costs and improves model capabilities across diverse tasks such as multi-turn dialogue and multilingual translation.
Key Points
- SLAP evaluates entire batch compositions for improved data selection efficiency.
- Utilizes distribution-aware stratified sampling for comprehensive data coverage.
- Achieves superior results with 20-40% less training data than traditional methods.
- Outperforms state-of-the-art techniques across various model architectures.
- Reduces computational costs while enhancing model performance.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2605. 23969v1 Announce Type: new Abstract: Instruction tuning has optimized the specialized capabilities of (LLMs), but it often requires extensive datasets and prolonged training times. The challenge lies in developing specific capabilities by identifying useful data and efficiently fine-tuning. High-quality and diverse pruned data can help models achieve lossless performance at a lower cost.
In this paper, we propose \textbf{SLAP}, a novel batch-aware data selection framework that evaluates the learnability of entire batch compositions rather than individual. SLAP ensures comprehensive data distribution coverage through distribution-aware stratified sampling while maximizing intra-batch diversity through relative distance optimization.
By leveraging Hessian-approximated gradient information for dynamic batch selection, SLAP significantly outperforms existing state-of-the-art methods across multiple model architectures (LLaMA, ChatGLM) and diverse downstream tasks including multi-turn dialogue, multilingual translation, and question answering.
Most notably, SLAP achieves superior performance with 20-40\% less training data compared to full dataset training, substantially reducing computational costs while maintaining or improving model capabilities. These results establish SLAP as a powerful approach for efficient and effective instruction tuning of large language models.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.