What problem does it solve?
Running TensorRT-LLM tests, evaluations, and benchmarks inside Docker requires manually launching containers, watching for hangs, and parsing logs to determine pass/fail status. This Skill automates that entire loop so an orchestrator can fire off a workload and receive a structured result without babysitting the process.
Core Features & Use Cases
- Background Docker Execution: Launches a fully-resolved
docker run command in the background with output tee'd to a log file, never blocking the foreground.
- Hang Detection and Recovery: Runs a monitoring loop that greps the log for hang indicators and kills the container automatically, reporting a
HANG_DETECTED status.
- Image Resolution and Build: Optionally detects GPU type and host CPU architecture, resolves the correct container image tag, and compiles the project before launching.
- Workflow-Aware Result Parsing: Parses logs differently for pytest, eval, custom, and benchmark workloads, returning a structured status report with exit code, summary, and errors.
- Use Case: An orchestrator needs to run a pytest suite for a TensorRT-LLM model on an H100 machine. It passes the Docker command, log path, and success patterns; the Skill builds the image, runs the tests, watches for hangs, and reports
PASSED with the pytest summary line.
Quick Start
Run this TensorRT-LLM pytest workload in local Docker using the provided docker command, monitor the log for hangs, and report whether the tests passed or failed.
