PyTorch, JAX, and compiler-level tools like Enzyme use three different representations for automatic differentiation—runtime tape, functional trace-and-transform, and source (IR) transformation—each w…
Google's TPU v5p and upcoming Ironwood TPU7x offer higher raw matrix throughput and simpler interconnects than NVIDIA H100 GPUs for large-scale AI training, but GPUs remain superior for inference with…
KerasFormers, an open-source library built entirely in Keras 3, launches with over 100 transformer models spanning vision, language, multimodal, and speech, supporting seamless execution on TensorFlow…
Shrijith Venkatramana, a developer building git-lrc, explains that NVIDIA Collective Communications Library (NCCL) is the critical infrastructure enabling multi-GPU training of large language models. …
AWS released a comprehensive guide for distributed AI training on its infrastructure, featuring reference architectures, test cases, and best practices for frameworks like PyTorch, Megatron-LM, NeMo, …
A developer debugging a JAX/Flax NNX training loop discovered that the loss was stuck at 10.82, indicating the model was performing no better than random guessing. The issue was traced to the training…
JAX's default_device context manager places arrays on the specified device but does not commit them, allowing JAX to move them to other devices. This caused array lookups to take over a second by trig…
Researchers released TycoonLE, a JAX-based reinforcement learning environment for long-horizon planning in a simulated logistics economy. The environment supports action legality, delayed rewards, and…
NVIDIA has introduced the NVFP4 training recipe in TransformerEngine, enabling 4-bit mixed-precision pre-training on Blackwell GPUs with no measurable accuracy loss compared to FP8 baselines. The reci…
A developer benchmarked quantum circuit simulation backends using a 20-qubit VQE workload on an NVIDIA RTX 5090 GPU, finding JAX/XLA 12.4x faster than PyTorch and 15.7x faster than TorchQuantum for po…
JAX defaults to loading data directly onto GPU memory when a CUDA-enabled version is installed, causing out-of-memory errors for large datasets that would fit in system RAM. The framework's `jax.devic…
Google released Magenta RealTime 2, an open-weights AI model and real-time inference engine that enables live, low-latency music generation on a MacBook. The model responds instantly to MIDI, audio, a…
A developer porting PyTorch LLM code to JAX using Flax encountered difficulties when attempting to store model checkpoints with Safetensors, as the library's Flax API expects flat dictionaries but Fla…
SpaceX has nearly completed version 1.0 of a proprietary AI training stack written in C, designed to exact-map to 220,000 GB300 processors with 800G network interface cards. The system, which leverage…
Google launched Coral, a full-stack platform for edge AI that combines an AI-first hardware architecture with a unified developer experience to enable efficient, local AI processing. The platform prov…
Researchers have released GenSBI, an open-source library for simulation-based inference built entirely in JAX, implementing flow matching, score matching, and denoising diffusion methods. The library …
AMD's ROCm (Radeon Open Compute platform) has reached production-ready maturity as an open-source alternative to NVIDIA's CUDA, now supporting LLM inference, fine-tuning, and image generation on AMD G…
TensorCircuit-NG, a quantum software stack built on AI infrastructure, treats quantum circuits as specialized tensor operations to leverage existing AI tooling for automatic differentiation, compilati…
MaxText has introduced new post-training capabilities, specifically Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), now available on single-host TPU configurations like v5p-8 and v6e-8. …
At the Google I/O conference, NVIDIA and Google Cloud announced new resources for their joint developer community of over 100,000 members, including learning paths for JAX on NVIDIA GPUs and an NVIDIA…