TurboGPT open-sourced: a 22KiB byte-level GPT trained in CUDA C++
What happened
On September 30, 2026, the Hacker News front page featured TurboGPT, a byte-level micro GPT training project written in CUDA C++ and open-sourced under the MIT license, able to train a 22KiB Transformer in 13 seconds. Public information so far covers only the project's technical implementation and training speed, with no follow-up reported.
Written by AI from the coverage · updated 3 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageShow HN: TurboGPT: train 22KiB transformer in 13s
TurboGPT 是一个用 CUDA C++ 实现的字节级微型 GPT 训练项目，采用 MIT 协议，可在 13 秒内训练 22KiB 的 Transformer。
Heat over time
Heat now 9·Comparable peak 9（Sep 30 06:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.