Skip to content

cli-reference.md

Latest commit

 

History

History
112 lines (86 loc) · 4.52 KB

cli-reference.md

File metadata and controls

112 lines (86 loc) · 4.52 KB

Complete CLI Reference

Complete reference for all command-line options in QuantDiff.

Main Script (main.py)

Required: Analysis Type

Exactly one of these flags must be specified:

Flag Description
--eval_flops Calculate FLOPs/MACs using forward hooks
--eval_sensitivity Analyze layer sensitivity to quantization
--optimize_mixed_precision Optimize mixed-precision configuration using Bang-for-Buck algorithm

General Options

Flag Type Default Description
--model_path str CompVis/stable-diffusion-v1-4 Hugging Face model identifier
--device choice auto Device: auto, cuda, cpu
--output_dir str results/ Output directory for JSON files
--output_name str None Custom output filename (without .json)
--no_save flag - Don't save JSON results
--quiet flag - Suppress detailed output

Sensitivity Analysis Options

Flag Type Default Description
--max_layers int None (all) Maximum layers to analyze
--sensitivity_bits int 4 Quantization bits for testing (4, 8, or 16)
--num_prompts int 100 Number of COCO prompts to use
--coco_path str None Path to COCO prompts file
--prompt_seed int 42 Random seed for prompt shuffling
--batch_size int 1 Batch size for generation

Mixed-Precision Optimization Options

Flag Type Default Description
--flops_file str results/flops_analysis/flops_analysis_unet.json Path to FLOPs analysis JSON file
--sensitivity_file str results/sensitivity_analysis/sensitivity_100_prompts.json Path to sensitivity analysis JSON file
--budget_multiplier float 0.5 Computational budget multiplier (0.0=all 4-bit, 0.5=medium, 1.0=all 16-bit)
--mixed_precision_config str None Path to existing mixed-precision config JSON to load instead of optimizing

Image Generation Script (generate_experiment.py)

Flag Type Default Description
--num_prompts int required Number of COCO prompts to use
--coco_path str None Path to COCO prompts file (default: prompts/coco_val2017.txt)
--model_path str CompVis/stable-diffusion-v1-4 Hugging Face model identifier
--device str cuda Device: cuda or cpu
--flops_file str results/flops_analysis/flops_analysis_unet.json Path to FLOPs analysis
--sensitivity_file str results/sensitivity_analysis/sensitivity_100_prompts.json Path to sensitivity analysis
--budget_levels float[] [0.1, 0.2, ..., 0.9] Budget multipliers (space-separated)
--seed int 42 Random seed for image generation
--prompt_seed int 42 Random seed for prompt selection
--experiment_dir str experiments Base directory for experiments

Evaluation Script (evaluate_experiment.py)

Flag Type Default Description
experiment_dir str required Path to experiment directory
--device str cuda Device: cuda or cpu
--save_results flag - Save results to CSV and JSON
--output_path str <experiment_dir>/evaluation_results.csv Custom output path

Usage Examples

Phase 1: FLOPs Analysis

python main.py --eval_flops
python main.py --eval_flops --device cuda
python main.py --eval_flops --output_dir custom_results/

Phase 2: Sensitivity Analysis

python main.py --eval_sensitivity
python main.py --eval_sensitivity --sensitivity_bits 8
python main.py --eval_sensitivity --num_prompts 50 --device cuda
python main.py --eval_sensitivity --max_layers 10 --batch_size 25

Phase 3: Mixed-Precision Optimization

python main.py --optimize_mixed_precision
python main.py --optimize_mixed_precision --budget_multiplier 0.3
python main.py --optimize_mixed_precision --flops_file custom/flops.json

Phase 4: Image Generation

python generate_experiment.py --num_prompts 50
python generate_experiment.py --num_prompts 100 --budget_levels 0.3 0.5 0.7
python generate_experiment.py --num_prompts 50 --device cuda --seed 123

Evaluation

python evaluate_experiment.py experiments/2025-10-31_14-30-45_50prompts/
python evaluate_experiment.py experiments/2025-10-31_14-30-45_50prompts/ --save_results