Rust tokenizer hits 8.27 GB/s, far outpacing Hugging Face tokenizers in parallel tests
deliprao · x · 2026-07-25
A benchmark on an Apple M4 Max shows a huge tokenizer throughput gap in parallel workloads.
- Gigatoken (Rust-optimized) reaches 8.27 GB/s on GPT-2 tokenization over a 12 GB dataset.
- tiktoken hits 61.5 MB/s.
- HF tokenizers comes in at just 6.2 MB/s.
The post argues that Rust-based tokenizers can behave dramatically better than Python-based stacks when parallel throughput matters.
More from Infra
- Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
- AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27
- Qwen3.8-27B hits 40 tok/s on Jetson Thor with INT4 + MTP, accuracy beats FP16 — No-Dragonfly6246 · 2026-10-27
- Celesto launches CI runners nearly 90% cheaper than GitHub Actions, with agent sandboxes — aniketmaurya · 2026-10-27
- Supabase founders on rebuilding the backend for a billion AI-built apps — Latent Space · 2026-10-27
- AW-0 runs at 230ms on a single RTX 5090, 2.4x faster than autoregressive code planning — trevordarrell · 2026-10-27