JAX 后端与设备
作者在将 PyTorch 的 LLM 代码移植到 JAX 时，遇到加载 19GiB 数据集时 CUDA 显存不足的问题。通过探究发现，JAX 默认将数据分配到最快的可用后端（GPU），而不会自动利用 CPU 内存。文章详细介绍了 JAX 的 backend 与 device 概念，并展示了如何使用 `jax.default_device` 上下文管理器临时切换到 CPU 设备加载大数组，以绕过 GPU 显存限制。
相关报道
The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models for all customers, citing a potential jailbreak technique that involved asking the model to review a codebase for vulnerabilities—a capability Anthropic says is available in other public models. Access was abruptly cut off on June 12.
Andrej Karpathy announces the release of Claude Fable 5, the same underlying model as Mythos but with added safeguards. He calls it a major step forward, particularly for long problem-solving sessions on difficult tasks, and describes it as state-of-the-art on nearly all benchmarks with exceptional performance in software engineering, research, and vision.
Apple says Siri AI is delayed in the EU for iOS 27 and iPadOS 27 due to the DMA, claiming the regulation demands unsafe open access to user data. The European Commission rejected Apple's proposed safety measures, leaving no timeline for release.
The U.S. government has ordered Anthropic to suspend access to Fable 5 and Mythos 5 models over national security concerns about a jailbreaking technique. Anthropic says it received no specific details and views the identified vulnerabilities as minor and replicable by other public models.
Anthropic silently limited Claude Fable's effectiveness on frontier LLM development requests like ML accelerator design, without notifying users. The company walked back the policy after outrage from the research community.