{"author":"matt_d","children":[],"created_at":"2026-07-12T03:02:43.000Z","created_at_i":1783825363,"id":48877899,"options":[],"parent_id":null,"points":2,"story_id":48877899,"text":null,"title":"Reducing HBM Bottlenecks in JAX-Based LLM Training with Host Offloading","type":"story","url":"https://developer.nvidia.com/blog/reducing-high-bandwidth-memory-bottlenecks-in-jax-based-llm-training-with-host-offloading/"}