Stop Using zstd! OpenZL Cuts AI Model Size by Half and Boosts Speed 10×
1 months ago 高效码农Stop Using zstd for Model Checkpoints! Meta’s OpenZL Cuts Size by Half and Runs 10× Faster Same CSV, zstd: 100 MB → OpenZL: 45 MB, and decompression is faster. Not keynote fluff—this is the real Grafana shot from Meta’s Nimble warehouse on launch day. 1. The 3 a.m. Page That Started It All Wednesday, 03:14. PagerDuty: “HDFS < 10 % free.” Ops adds 2 PB—buys two weeks. Every shard is already at zstd -19; going to level 22 will only turn GPUs into expensive space-heaters. Meta’s compression team shipped OpenZL instead. Same data, two weeks later: –18 % disk, –5 …
ParScale Parallel Computing: The Third Paradigm Revolutionizing AI Scaling
6 months ago 高效码农The Third Paradigm of AI Scaling: Demystifying ParScale’s Parallel Computing Revolution Introduction: Shattering the “Impossible Trinity” of Language Models The AI community has long struggled with balancing three critical factors: model performance, computational cost, and deployment efficiency. Traditional approaches force painful tradeoffs: ◉ Parameter Scaling: While increasing parameters boosts capability, it incurs exponential costs (GPT-3’s training consumed energy equivalent to 126 Danish households annually) ◉ Inference Optimization: Compression techniques like knowledge distillation often sacrifice up to 73% of model effectiveness The groundbreaking 2025 study Parallel Scaling Law for Language Models introduces a third way – ParScale parallel scaling. This China-led …