End-to-End On-Device Quantization-Aware Training for LLMs at Inference Cost
cs.LG, cs.AI
Submitted: 2025-08-21
Updated: 2026-11-05
Terminology
Sources
- Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
- DeepZero: Scaling up Zeroth-Order Optimization for Deep Model Training
- EfficientQAT: Efficient Quantization-Aware Training for Large Language Models
- Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
- Model Agnostic Contrastive Explanations for Structured Data
- DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs
- Learned Step Size Quantization
- GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
- LoftQ: LoRA-Fine-Tuning-Aware Quantization for Large Language Models
- BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction
- RoBERTa: A Robustly Optimized BERT Pretraining Approach
- LLM-QAT: Data-Free Quantization Aware Training for Large Language Models
- Pointer Sentinel Mixture Models
- Accurate LoRA-Finetuning Quantization of LLMs via Information Retention
- SQuAD: 100,000+ Questions for Machine Comprehension of Text
- Fine-tuning Quantized Neural Networks with Zeroth-order Optimization
- OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models
- CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledge
- Perturbation-efficient Zeroth-order Optimization for Hardware-friendly On-device Training
- Gemma 3 Technical Report
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis