LLM fine-tuning

federated learning

Sparse-Orthogonal LoRA Enables Wireless Federated LLM Training

New research introduces SO-LoRA, combining sparse and orthogonal low-rank adaptation to enable efficient multi-task LLM fine-tuning over wireless networks with reduced interference.

LLM fine-tuning

Influence-Preserving Proxies Accelerate LLM Fine-Tuning Data Sele

New research introduces proxy methods that preserve gradient influence signals while dramatically reducing computational costs for selecting optimal training data in large language model fine-tuning.

LLM fine-tuning

Zero-Order Optimization Enables Memory-Efficient LLM Fine-Tuning

New research introduces learnable direction sampling for zero-order optimization, dramatically reducing memory requirements for fine-tuning large language models without sacrificing performance.

LoRA

Study Finds Vanilla LoRA Matches Complex Variants With Proper Tun

New research reveals that standard LoRA fine-tuning can achieve performance comparable to sophisticated variants when learning rates are properly optimized, challenging assumptions about adapter complexity.

LLM fine-tuning

RVPO: Stabilizing LLM Fine-Tuning Through Variance Control

New research introduces Ratio-Variance Regularized Policy Optimization (RVPO), a method that stabilizes reinforcement learning from human feedback by controlling importance sampling variance in LLM training.

LLM fine-tuning

Chronicals Framework Achieves 3.51x LLM Fine-Tuning Speedup

New open-source framework Chronicals claims significant performance gains over popular fine-tuning tool Unsloth, promising faster and more efficient LLM training for researchers and developers.