Deepity releases C++/CUDA predictive coding library, edges JAX baseline on MNIST

A developer released Deepity, a from-scratch predictive coding engine in C++20 with hand-written AVX2/AVX-512 kernels, a CUDA backend and Python bindings. In matched MNIST benchmarks it reached 98.21% accuracy in 487.5s, beating jpc's 97.55% and ngc-learn's 98% in 1707.7s.

• Raschka releases from-scratch reasoning model video course2026-10-26
• Developer proposes speculative decoding using 2-bit draft plus RAM residual2026-10-22
• Hugging Face adds native TorchTPU PyTorch backend2026-10-20
• PyTorch 2.15 brings FP8 and faster decoding to Mac MPS2026-10-20
• Open GPU kernel engineering series covers GEMM to FlashAttention2026-10-19