Mark Johnson
Passionate web designer and art lover.
15.05.2020
Microsoft Builds ZeRO + DeepSpeed — Training Trillion-Parameter Models
Microsoft releases DeepSpeed and ZeRO, optimizer technologies that make training trillion-parameter models feasible. The groundwork for GPT-3-scale training is laid.
0 Comments1 Minute
15.06.2020
OpenAI Releases GPT-3 — 175 Billion Parameters Change Everything
OpenAI releases GPT-3 with 175B parameters via private API. Developers unlock previously impossible use cases overnight. A new era of LLM products begins.
0 Comments1 Minute
15.07.2020
NVIDIA A100 Ampere GPU Ships — AI Compute Leaps Forward
NVIDIA ships the A100 Ampere GPU with 312 TFLOPS of FP16 compute and 40GB HBM2. It becomes the workhorse of every hyperscale AI training cluster.
0 Comments1 Minute