How kainam15/ac-prof works
AC-Prof is a model-inference benchmarking application: users select a model and resource/input configuration, prepare a containerized runtime, execute measurement cases, and inspect reproducible results. The graph retains documented collection modes, workload families, model coverage, post-hoc profiling, and comparison/reporting capabilities. Several runtime links are based on README descriptions because the relevant implementation excerpts were not provided; those edges are marked dashed with null evidence.
Main entry points: orchestrator.py Run orchestration · client.py Measurement client · server.py Inference server · artifact_layout.py Experiment artifacts
Components 25
User interfaces
- Terminal UI
acprof/tui/app.py - CLI commands
acprof/cli/main.py
Experiment planning
- Model resolution
acprof/model_resolution.py - Model contracts
acprof/model_spec.py - Model coverage
acprof/host/model_coverage.py - Host capabilities
acprof/host/detect.py - Shared model store
acprof/host/model_store.py - Runtime images
acprof/host/image_management.py
Inference and measurement
- Run orchestration
acprof/host/orchestrator.py - Run state and resume
acprof/host/run_state.py - Inference server
acprof/container/server.py - Task workloads
acprof/workloads/contract.py - Measurement client
acprof/host/client.py - Metric definitions
acprof/metric_registry.py - Optional profilers
acprof/host/profilers - Energy monitoring
acprof/monitors/energy_cpu.py - Packet latency
acprof/packet/merge_packet_latency.py
Results and analysis
- Experiment artifacts
acprof/artifact_layout.py - CSV measurements
acprof/result_csv.py - Post-hoc profiling
acprof/host/posthoc/service.py - Comparison and analysis
acprof/analysis/comparison.py - Reports and plots
acprof/plotting/report.py
Other components
- Benchmark user
- Docker Engine
- Hugging Face Hub
Connections 28
- Benchmark user → Terminal UI: starts runs
README.md - Benchmark user → CLI commands: configures runs
README.md - Terminal UI → Run orchestration: launches run
- CLI commands → Run orchestration: launches run
- Run orchestration → Host capabilities: checks host
README.md - Run orchestration → Model resolution: resolves model
- Model resolution → Model contracts: uses contract
acprof/model_resolution.py - Run orchestration → Shared model store: prepares weights
README.md - Hugging Face Hub → Shared model store: provides weights
README.md - Run orchestration → Runtime images: prepares runtime
README.md - Runtime images → Docker Engine: builds or pulls
README.md - Run orchestration → Run state and resume: tracks cases
- Run orchestration → Inference server: starts inference
README.md - Inference server → Task workloads: serves task
acprof/container/server.py - Measurement client → Inference server: sends requests
README.md - Measurement client → Metric definitions: records metrics
acprof/host/client.py - Measurement client → Energy monitoring: collects energy
README.md - Measurement client → Optional profilers: runs optional profiling
README.md - Measurement client → Packet latency: merges packet latency
acprof/packet/merge_packet_latency.py - Run orchestration → Experiment artifacts: publishes results
README.md - Measurement client → CSV measurements: writes measurements
- Experiment artifacts → CSV measurements: stores measurements
- Benchmark user → Model coverage: assesses coverage
README.md - Benchmark user → Post-hoc profiling: requests profiling
README.md - Post-hoc profiling → Experiment artifacts: updates artifacts
acprof/host/posthoc/service.py - Benchmark user → Comparison and analysis: compares results
README.md - CSV measurements → Comparison and analysis: supplies measurements
- Comparison and analysis → Reports and plots: generates reports
README.md