Total time spent in proxy (answer_receive_end_at - start_time).

Store translation in cache.

Helper: Render status header

Helper: Render single service line

Run benchmark with specified concurrency. Load modes: - fixed: maintain fixed number of active requests - qps: emit requests at fixed

Run the benchmark with specified concurrency using threads If target_langs is provided, it should be a list with one target language per chunk.

Run inference server in Docker and return the process.

Build Docker command for sglang.

Build Docker command for vllm.

Run evaluation for multiple tasks × multiple models. Returns: all_results: {model_name: {task_key: metrics}}

Run batch evaluation on all pairs and return results.

Run health-monitor remotely and health-client locally to test health monitoring via vsock.

Run evaluation for a single model on a single task. Returns: outputs: Model outputs (parallel to samples) stats: Execution st

Unified proxy test function supporting both CLI args and config files. Architecture: - Local node: Always runs SOCKS5 proxy (port 8116)

Run seal-server locally and seal-client remotely, printing client output.

Run a single benchmark iteration.

Explicitly save cache.

Save benchmark results to CSV in the same format as the logging system.

Helper: trim and split CSV line

Split text into chunks of approximate length.

Let's simplify attestation api

Return cache statistics.