HiFloat4 Format for Language Model Pre-training on Ascend NPUs
cs.LG, cs.AI, cs.CL
Submitted: 2026-04-09
Updated: 2026-09-25
Terminology
Sources
- Pretraining Large Language Models with NVFP4
- Metis: Training LLMs with FP4 Quantization
- Quartet: Native FP4 Training Can Be Optimal for Large Language Models
- INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats
- Oscillation-Reduced MXFP4 Training for Vision Transformers
- RaZeR: Pushing the Limits of NVFP4 Quantization with Redundant Zero Remapping
- Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats
- FP4 All the Way: Fully Quantized Training of LLMs
- Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling
- Elucidating the Design Space of FP4 training
- Scaling Laws for Neural Language Models
- DeepSeek-V3 Technical Report
- HiFloat4 Format for Language Model Inference
- FP8 Formats for Deep Learning
- Quartet II: Accurate LLM Pre-Training in NVFP4 by Improved Unbiased Gradient Estimation
- Microscaling Data Formats for Deep Learning
- Training LLMs with MXFP4
- Optimizing Large Language Model Training Using FP4 Quantization
- Scalable Training of Mixture-of-Experts Models with Megatron Core
- Accurate INT8 Training Through Dynamic Block-Level Fallback
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks