Tools
fun hack, GPT-2 inference running on my jailbroken Kindle Scribe this is real speed, like ~13 tok/s at 4-bit quant of the 124M sized model
Lucas Gelfond (@gucaslelfond) demonstrated that GPT‑2 inference can be run on a jailbroken Kindle Scribe, achieving roughly 13 tokens per second with a 4‑bit quantized model of the 124 M‑parameter var
Lucas Gelfond (@gucaslelfond) demonstrated that GPT‑2 inference can be run on a jailbroken Kindle Scribe, achieving roughly 13 tokens per second with a 4‑bit quantized model of the 124 M‑parameter variant. The demo was shared publicly on October 25, 2026 and highlights a practical deployment of low‑precision language models on unconventional hardware.
Related
- Inception Mercury 2 is live on AI Gateway
- How fast is 10 tokens per second really?
- Multi-token Residual Prediction
Source: Linus Lee (X) | 2026-10-25