Wazafak — Benchmarks

Gigatoken is a new open-source tokenizer that is ~1000x faster than HuggingFace's tokenizers and ~100x faster than OpenAI's tiktoken for most tokenizer definitions on most machines.

A recent study measures input latency on Linux, comparing X11 and Wayland, with Variable Refresh Rate (VRR) and DXVK low-latency fork. The results show that X11 has lower latency than Wayland, but the difference is small. VRR has the biggest effect, reducing latency by 0.26 to 0.45 ms.

Benchmarking 15 decommissioned NVIDIA enterprise GPUs with modern AI workloads reveals surprising performance and value, despite being EOL'd and less power efficient.