fishy.analysis.benchmark¶

Benchmarking utilities for model performance evaluation.

Functions

fishy.analysis.benchmark.get_peak_vram() → float[source]¶: Gets peak VRAM usage in MB (if CUDA is available).

fishy.analysis.benchmark.measure_inference_performance(model: Module, input_size: tuple, device: device | str, num_iterations: int = 100) → Dict[str, float][source]¶: Measures inference latency and throughput.

fishy.analysis.benchmark.measure_model_size(model: Module) → float[source]¶: Calculates the model size in Megabytes.

fishy.analysis.benchmark.run_benchmark(model: Any, input_dim: int, device: device, ctx: Any, training_time: float | None = None) → Dict[str, Any][source]¶: Runs a full benchmark suite and saves results.

s