Really neat local vision loop on a MacBook! 🦖💻⚡ A 12B decision model fine-tuned from Gemma 4 ("Jev-Omni") plays Chrome Dino in real time: 100ms screen slices, 29 new cached tokens/frame, and 10 decisions/sec locally. Weights: https://huggingface.co/akhilaaa3/Jev-Omni

Show Post

🐍🖼️ Pillow 13.0.0 is out! 🔒 Fixes for 38 security issues, including 2 critical. Please upgrade! ⚡ Import is ~2x faster, getbbox() up to 24x, transpose() up to 10x, and JPEG 2000, TIFF, text and ImageCms up to 3x in macOS/Linux wheels 🆕 Magic Kernel Sharp resize filters 🐍 Python 3.15 and PyPy 3 […]

Show Post

Who wants smaller Zed binaries? In the next Stable release, Zed's executable drops by ~30% across every platform.

Show Post

Just released ryl v0.25: the fast Rust YAML linter is now a formatter too! Our biggest release yet, with over 140 merged PRs. Try `ryl format`. Thanks to cleder and psthindal for the requests! ryl-docs.pages.dev/formatter/ #rust #yaml

Show Post

The recent Tokenizers v1 update seriously sped up tokenization for any NLP task, but now @SimonBrandeis is cooking another improvement, especially for BERT tokenizers: an additional ~5x (or ~50x total from the previous tokenizers v0.23).

Show Post