A 1B-parameter hierarchical recurrent model pretrained on 40B instruction-response tokens achieves 60.7% MMLU and strong results on ARC-C, DROP, GSM8K, and MATH while using 100-900x fewer tokens than standard baselines.
Neural gradients are near-lognormal: Improved quantized and sparse training
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
years
2026 3verdicts
UNVERDICTED 3representative citing papers
Introduces adaptive unbiased quantization algorithms with provable guarantees that preserve inner products and yield 2-10x faster practical methods for adaptive stochastic quantization.
Deployment-aligned low-precision NAS recovers about two-thirds of the accuracy drop from post-training quantization, achieving 0.826 mIoU on-device for a 95k-parameter model on Intel Movidius Myriad X without added complexity.
citing papers explorer
-
HRM-Text: Efficient Pretraining Beyond Scaling
A 1B-parameter hierarchical recurrent model pretrained on 40B instruction-response tokens achieves 60.7% MMLU and strong results on ARC-C, DROP, GSM8K, and MATH while using 100-900x fewer tokens than standard baselines.
-
Inner Product Aware Quantization: Provably Fast, Accurate, and Adaptive Algorithms
Introduces adaptive unbiased quantization algorithms with provable guarantees that preserve inner products and yield 2-10x faster practical methods for adaptive stochastic quantization.
-
Deployment-Aligned Low-Precision Neural Architecture Search for Spaceborne Edge AI
Deployment-aligned low-precision NAS recovers about two-thirds of the accuracy drop from post-training quantization, achieving 0.826 mIoU on-device for a 95k-parameter model on Intel Movidius Myriad X without added complexity.