从零编写LLM,第34a部分——为LLM训练构建JAX训练循环
作者以Sebastian Raschka的《从零构建大语言模型》为教材学习现代AI,在独立完成PyTorch模型训练后,为了验证理解而非机械复现代码,选择用JAX框架编写LLM训练循环。本文记录了从零构建A-to-A(输入等于输出)的最小模型与训练流程的过程,涉及数据加载、设备内存管理、优化器配置等关键技术细节,以及从PyTorch迁移到JAX/Flax NNX/Optax生态的经验教训。
作者以Sebastian Raschka的《从零构建大语言模型》为教材学习现代AI,在独立完成PyTorch模型训练后,为了验证理解而非机械复现代码,选择用JAX框架编写LLM训练循环。本文记录了从零构建A-to-A(输入等于输出)的最小模型与训练流程的过程,涉及数据加载、设备内存管理、优化器配置等关键技术细节,以及从PyTorch迁移到JAX/Flax NNX/Optax生态的经验教训。
Max Weinbach says he had early access to OpenAI's new model GPT-5.6 Sol, calling it his favorite model by far. He highlights that it never gives up and will keep reasoning until it's done. OpenAI announced that GPT-5.6 Sol, along with Terra and Luna, will launch publicly on Thursday, with preview access expanding globally now.
OpenAI introduced GPT‑Live, a new version of its AI model designed for real-time, interactive voice conversations. The model enables more natural, low-latency spoken interactions with users, marking a step toward more conversational AI capabilities.
xAI has released Grok 4.5, a new version of its AI model, which the company claims is its most advanced yet.
Cognition has released SWE-1.7, a new version of its AI coding agent that they claim approaches near GPT-5.5 and Opus levels of intelligence, marking a significant advancement in autonomous software engineering capabilities.
Newer Claude models sometimes invent extra keys in tool call arguments, breaking validation in Pi's edit tool. The author suspects post-training for Claude Code's forgiving harness makes alternative schemas fail. This suggests closed RL training can degrade general tool-use reliability.