How to Run Turing-Complete Malware in PyTorch's Backward Pass

Elon Litman demonstrates how a malicious PyTorch loss function can report a value of zero while secretly manipulating model weights through the backward pass. By exploiting differences between floating-point arithmetic and automatic differentiation, he shows how gradient descent can execute arbitrary computations inside a neural network, creating a potential attack vector for AI training...

https://elonlit.com/scrivings/how-to-run-turing-complete-malware-in-pytorchs-backward-pass/

lightseekorg / tokenspeed

TokenSpeed is a speed-of-light LLM inference engine.

https://github.com/lightseekorg/tokenspeed

django-lifecycle - hooks for Django models (Signals alternative)

Learn how to usedjango-lifecycleas an alternative to Django signals, with declarative hooks that run when models are created, updated, deleted, or saved. The package also supports conditional hooks that trigger only when specific fields change or transition between particular values.

https://www.youtube.com/watch?v=sNdopNFOEB8

Decoding base64 in Python at 10 gigabytes per second

Daniel Lemire introduces fastpybase64, a Python library built on SIMD-accelerated simdutf that provides a drop-in alternative to Python's standard base64 module. In single-threaded benchmarks decoding Base64 data containing line breaks, it achieves over 13 GB/s throughput, approximately 27 times faster than the standard library and four times faster than pybase64.

https://lemire.me/blog/2026/10/17/decoding-base64-in-python-at-10-gigabytes-per-second/

Introducing TorchNeuron: Native PyTorch backend for AWS Trainium

PyTorch and AWS introduce TorchNeuron, an open-source backend that lets developers run PyTorch models natively on AWS Trainium2 and Trainium3 accelerators with minimal code changes. Now available in public beta, it supports eager execution, torch.compile, distributed training, custom kernels through the Neuron Kernel Interface (NKI), and performance profiling using familiar PyTorch APIs.

https://pytorch.org/blog/introducing-torchneuron-native-pytorch-backend-for-aws-trainium/

Introducing TorchTPU

TorchTPU is an open source collaboration between Google and Meta that introduces a native and principled PyTorch backend for Google TPUs. It enables seamless out-of-the-box model execution and TPU migration while delivering industry-competitive inference and training performance through eager execution optimizations, torch.compile integration, and native hardware acceleration.

https://pytorch.org/blog/introducing-torchtpu/

SQLBuild

SQL transformation framework with custom compile-time checks, end-to-end testing and safe renames that keep your table history.

https://github.com/chio-labs/sqlbuild