How kinsingo/ai-microbmt works
AI-MicroBMT is a benchmark-reproduction toolkit: it converts AI-BMT exports into organized result CSVs, computes Unified Deployment Scores and use-case rankings, and analyzes benchmark variants to produce figures, summaries, and LaTeX tables. Its accelerator-specific scripts are reference implementations for collecting device benchmarks, not part of the Python analysis runtime. The README documents the main workflow; source-index references confirm the analysis scripts use shared utilities and configuration. Some individual analysis and accelerator implementation details were not sampled.
Main entry points: mobilenet variant Variant result CSVs · analysis_charts Charts and tables · analyze_results_singleStream_offline.py Scenario comparison · analyze_activation_sweep.py Activation sweep
Components 16
Benchmark Collection
- Exported benchmark CSV
mobilenet variant - CSV converter
convert_and_evaluate.py - Variant result CSVs
mobilenet variant
UDS Scoring
- UDS sub-scores
1. Create UDS Scores.py - Weighted rankings
2. UDS cases.py
Analysis
- Analysis settings
analysis_config.py - Data utilities
utils.py - Scenario comparison
analyze_results_singleStream_offline.py - Activation sweep
analyze_activation_sweep.py - Resolution latency
analyze_results_input_resolution_singleStream.py - Resolution throughput
analyze_results_input_resolution_offline.py - Case analysis
generate_cases_analysis.py - Charts and tables
analysis_charts
Accelerator References
- Device evaluation scripts
accelerator_AIBMT_evalution_scripts
Other components
- Benchmark operator
- AI-BMT application
Connections 26
- Benchmark operator → AI-BMT application: runs benchmarks
README.md - AI-BMT application → Exported benchmark CSV: exports results
README.md - Benchmark operator → CSV converter: runs conversion
README.md - Exported benchmark CSV → CSV converter: reads
README.md - CSV converter → Variant result CSVs: writes variants
README.md - Variant result CSVs → UDS sub-scores: scores data
README.md - UDS sub-scores → Weighted rankings: provides sub-scores
README.md - Variant result CSVs → Scenario comparison: analyzes results
README.md - Variant result CSVs → Activation sweep: analyzes variants
README.md - Variant result CSVs → Resolution latency: analyzes sweep
README.md - Variant result CSVs → Resolution throughput: analyzes sweep
README.md - Weighted rankings → Case analysis: classifies cases
- Scenario comparison → Charts and tables: generates charts
README.md - Activation sweep → Charts and tables: generates charts
README.md - Resolution latency → Charts and tables: generates charts
README.md - Resolution throughput → Charts and tables: generates charts
README.md - Case analysis → Charts and tables: writes tables
README.md - Scenario comparison → Data utilities: imports
analyze_results_singleStream_offline.py - Scenario comparison → Analysis settings: imports
analyze_results_singleStream_offline.py - Activation sweep → Data utilities: imports
analyze_activation_sweep.py - Activation sweep → Analysis settings: imports
analyze_activation_sweep.py - Resolution latency → Data utilities: imports
analyze_results_input_resolution_singleStream.py - Resolution latency → Analysis settings: imports
analyze_results_input_resolution_singleStream.py - Resolution throughput → Data utilities: imports
analyze_results_input_resolution_offline.py - Resolution throughput → Analysis settings: imports
analyze_results_input_resolution_offline.py - Benchmark operator → Device evaluation scripts: uses as reference
README.md