h100-sglang-diffusion

Run diffusion smoke tests on an H100 by connecting to the remote host, entering the sglang_bbuf container, and executing standardized validation steps within the prepared environment. This workflow ensures safe and repeatable experiments.

Validate Triton and CUDA kernels remotely by accessing the H100 SGLang setup. The workflow automates kernel checks within the sglang_bbuf container, ensuring consistent results without local hardware constraints.

Execute torch.compile diffusion experiments on SGLang by connecting to the h100_sglang host and navigating to the repository at /data/bbuf/repos/sglang. The container context ensures proper environment variable handling.

Local H100 hardware is not required. SSH into the prepared remote H100 environment to perform diffusion development and validation, enabling safe experiments without local hardware constraints.

Secure environment variables during remote GPU validation by using standardized isolation workflows within the sglang_bbuf container. This approach protects local worktrees and secrets while ensuring repeatable steps.