Picture for Moran Shkolnik

Moran Shkolnik

EXAQ: Exponent Aware Quantization For LLMs Acceleration

Add code
Oct 04, 2024
Figure 1 for EXAQ: Exponent Aware Quantization For LLMs Acceleration
Figure 2 for EXAQ: Exponent Aware Quantization For LLMs Acceleration
Figure 3 for EXAQ: Exponent Aware Quantization For LLMs Acceleration
Figure 4 for EXAQ: Exponent Aware Quantization For LLMs Acceleration
Viaarxiv icon

DropCompute: simple and more robust distributed synchronous training via compute variance reduction

Add code
Jun 18, 2023
Viaarxiv icon

Neural gradients are lognormally distributed: understanding sparse and quantized training

Add code
Jun 17, 2020
Figure 1 for Neural gradients are lognormally distributed: understanding sparse and quantized training
Figure 2 for Neural gradients are lognormally distributed: understanding sparse and quantized training
Figure 3 for Neural gradients are lognormally distributed: understanding sparse and quantized training
Figure 4 for Neural gradients are lognormally distributed: understanding sparse and quantized training
Viaarxiv icon

Robust Quantization: One Model to Rule Them All

Add code
Feb 18, 2020
Figure 1 for Robust Quantization: One Model to Rule Them All
Figure 2 for Robust Quantization: One Model to Rule Them All
Figure 3 for Robust Quantization: One Model to Rule Them All
Figure 4 for Robust Quantization: One Model to Rule Them All
Viaarxiv icon

Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks

Add code
Sep 17, 2019
Figure 1 for Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
Figure 2 for Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
Figure 3 for Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
Figure 4 for Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
Viaarxiv icon