What problem does it solve? Measuring text-to-video generation performance on NVIDIA GPUs requires manual container setup, script orchestration, and log parsing. This Skill automates the full Open-Sora v2 inference evaluation workflow, from Docker launch to structured frames_per_second metrics. ## Core Features & Use Cases - Automated Container Orchestration: Launches the prebuilt Docker image with correct GPU capabilities, volume mounts for code, weights, and logs. - Configurable Inference Runs: Supports 256px and 768px resolutions, custom prompts, and offload toggling via environment variables, fixed to single-GPU execution. - Structured Metric Extraction: Parses inference logs for seconds-per-iteration, output video paths, and peak CUDA memory, then writes aggregated results to eval_result.json. - Use Case: Ask the agent to evaluate Open-Sora v2 text-to-video inference at 768px on an NVIDIA GPU, and receive a frames_per_second score with memory usage and the generated video path. ## Quick Start Ask the agent to run an Open-Sora v2 text-to-video inference benchmark on NVIDIA at 256px and report the frames per second.
