HOPE: A New Method for Pruning Mixture-of-Experts Models Without Losing Performance
Hugging Face introduces Higher-Order Pruning of Experts (HOPE), a technique that accounts for expert interactions when removing redundant parameters from MoE language models.
Hugging Face Research Cuts Knowledge Distillation Memory by 80%, Enabling Single-GPU Training
New offline logits caching and fused KL loss reduce VRAM overhead from 250GB to practical single-GPU levels, opening large-scale model compression to resource-constrained teams.
PP-OCRv6 Scales Multilingual Text Detection to 50 Languages Across 1.5M–34.5M Parameters
PaddleOCR's latest model family achieves 86.2% detection accuracy in medium tier, with three deployment tiers optimized for edge to server-side workloads.