All news

ThinkingBox turns database end-state checks into RL rewards for post-training agents, with training code released.

Part 1 of a guide to system-level profiling of PyTorch inference on AWS Trainium.

Specializing the tokenizer vocabulary of a fine-tuned small model cut token counts and inference cost by 77% in a production essay grader.

TorchTPU is a native PyTorch backend for TPUs, so Hugging Face models run on TPU without torch_xla or JAX.

PyTorch 2.15 brings FP8 and faster decoding kernels to the Apple MPS backend.

IBM releases granite-speech-5.0-2b and updated smaller speech models for transcription, translation and speaker attribution.

NVIDIA Kumo Relational is a pretrained relational foundation model that predicts churn or fraud directly from database tables with no training.

LitPivot helps researchers adjust ideas by placing them in the related literature as they read.

IBM shows ensemble forecasting across four Granite time-series models behind one interface.