谷歌新发布的Colab CLI让开发者和AI代理能从终端在远程Colab GPU和TPU上运行Python
英文摘要
Google released the Colab CLI, a command-line interface that connects local terminals to remote Colab runtimes. It allows developers and AI agents to run code on cloud GPUs and TPUs without leaving the terminal. The tool is open-source under Apache 2.0 and includes a bundled skill file for agent integration. It supports provisioning GPUs like T4, A100, and TPUs, and demonstrates a fine-tuning pipeline for Gemma 3 1B. The CLI is designed for scripted, automated, and agent-driven workflows.
中文摘要
Google发布了Colab CLI，这是一个命令行界面，将本地终端连接到远程Colab运行时。它允许开发者和AI代理在不离开终端的情况下在云端GPU和TPU上运行代码。该工具采用Apache 2.0开源许可，并附带了供代理使用的技能文件（COLAB_SKILL.md）。它支持配置T4、A100等GPU以及TPU，并演示了针对Gemma 3 1B模型的微调流程。该CLI专为脚本化、自动化和代理驱动的工作流设计。
关键要点
Colab CLI connects the local terminal to remote Colab runtimes for GPU and TPU computing.
Colab CLI将本地终端连接到远程Colab运行时，用于GPU和TPU计算。
It is designed for both developers and AI agents, such as Claude Code, Codex, and Antigravity.
它是为开发者和AI代理（如Claude Code、Codex和Antigravity）设计的。
The tool is open-source under Apache 2.0 and includes a bundled COLAB_SKILL.md for agent context.
该工具采用Apache 2.0开源许可，并附带COLAB_SKILL.md供代理使用。
Key commands include colab new (with --gpu/--tpu flags), colab exec, colab stop, colab download, and colab log.
关键命令包括colab new（带--gpu/--tpu标志）、colab exec、colab stop、colab download和colab log。
An example shows fine-tuning Gemma 3 1B using QLoRA on a T4 GPU, all driven by an AI agent.
一个示例展示了在T4 GPU上使用QLoRA微调Gemma 3 1B模型，整个过程由AI代理驱动。