ZARZA We are Zarza, the prestigious firm behind major projects in information technology.

HashAttention: Semantic Sparsity for Faster Inference

Mechanical Dreams

16/01/2025 11 min Temporada 1 Episodio 33

HashAttention: Semantic Sparsity for Faster Inference

Listen "HashAttention: Semantic Sparsity for Faster Inference"

Descargar episodio

More episodes of the podcast Mechanical Dreams

DeepSeek OCR 24/10/2025

The Coverage Principle - How Pre-training Enables Post-Training 23/10/2025

Continual Learning via Sparse Memory Finetuning 22/10/2025

Untitled Episode 10/10/2025

Characterization and Mitigation of Training Instabilities in Microscaling Formats 08/10/2025

Demystifying Synthetic Data in LLM Pre-training- A Systematic Study of Scaling Laws, Benefits, and Pitfalls 07/10/2025

Data Mixing Can Induce Phase Transitions in Knowledge Acquisition 06/10/2025

Apertus Tech Report 21/09/2025

Learning Facts at Scale with Active Reading 20/09/2025

Fantastic Pretraining Optimizers and Where to Find Them 19/09/2025

Ver todos los episodios