PulseAugur / Brief
EN
LIVE 09:20:00

Brief

last 24h
[2/2] 221 sources

Multi-source AI news clustered, deduplicated, and scored 0–100 across authority, cluster strength, headline signal, and time decay.

  1. RESEARCH · Together AI blog English(EN) · · [2 sources]

    FlashAttention

    Together AI has released FlashAttention-3 and FlashAttention-4, significant upgrades to their GPU-accelerated attention mechanism for large language models. FlashAttention-3, designed for Hopper GPUs, achieves up to 75% utilization and 1.5-2x speedup over its predecessor by exploiting new hardware features like Tensor Cores and Tensor Memory Accelerator, and supporting FP8 precision. FlashAttention-4, optimized for Blackwell GPUs, further enhances performance by pipelining computations and addressing bottlenecks in transcendental functions and memory traffic, reaching 71% utilization and offering substantial speedups over existing libraries. AI

    FlashAttention

    IMPACT These optimized attention mechanisms promise significantly faster LLM training and inference, enabling longer context windows and more efficient GPU utilization.

  2. SIGNIFICANT · Together AI blog English(EN) · · [3 sources]

    Salesforce, Zoom, InVideo Train Faster with Together AI Turbocharged with NVIDIA Blackwell

    Together AI has launched new GPU clusters featuring NVIDIA's Blackwell platform, offering significant speedups for AI training and inference. These clusters, powered by the Together Kernel Collection, achieve up to 90% faster training speeds compared to previous NVIDIA H100 hardware, processing over 15,000 tokens per second for large models. Early access customers like Salesforce and Zoom have reported substantial performance gains, with some experiencing double the training speed. Together AI's optimization efforts span custom kernels, inference engines, and speculative decoding, aiming to redefine efficiency in AI model development and deployment. AI

    Salesforce, Zoom, InVideo Train Faster with Together AI Turbocharged with NVIDIA Blackwell

    IMPACT Accelerates AI training and inference, potentially lowering costs and increasing the pace of model development and deployment for enterprises.