exec-slurm-compile

What problem does it solve?

This Skill streamlines compiling TensorRT-LLM on a remote SLURM cluster by coordinating container preparation, job submission, live monitoring, and build verification so you don’t need to compile directly on your local machine.

Core Features & Use Cases

• SLURM batch compilation workflow: Submits a containerized compilation job using sbatch and runs the build inside an enroot container.
• enroot container pre-dumping: Optionally pre-imports the Docker image into a .sqsh image via a dedicated SLURM job to reduce compile-time download overhead.
• Live monitoring and log-based verification: Actively polls SLURM status, tails the srun log, and validates success via SLURM exit code and build log messages.
• Use Case: You need to build TensorRT-LLM for a specific GPU architecture on a cluster partition with the right TensorRT environment packaged in an NVIDIA container.

Quick Start

Use the enroot-import step to generate a .sqsh from the TensorRT-LLM Docker image, then submit the compilation with sbatch using that .sqsh, your TensorRT-LLM repo path, and your SLURM partition and account.