utils/benchmark_utils.py

import argparse
import functools
import os
from torch.profiler import record_function


def create_parser():
    """Creates CLI args parser."""
    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)

    # general options
    parser.add_argument("--ckpt", type=str, default="black-forest-labs/FLUX.1-schnell",
                        help="Model checkpoint path")
    parser.add_argument("--prompt", type=str, default="A cat playing with a ball of yarn",
                        help="Text prompt")
    parser.add_argument("--cache-dir", type=str, default=os.path.expandvars("$HOME/.cache/flux-fast"),
                        help="Cache directory for storing exported models")
    parser.add_argument("--use-cached-model", action="store_true",
                        help="Attempt to use cached model only (don't re-export)")
    parser.add_argument("--device", type=str, choices=["cuda", "cpu"], default="cuda",
                        help="Device to use")
    parser.add_argument("--num_inference_steps", type=int, default=4,
                        help="Number of denoising steps")
    parser.add_argument("--output-file", type=str, default="output.png",
                        help="Output image file path")
    parser.add_argument("--seed", type=int, default=42,
                        help="Random seed to use")
    # file path for optional output PyTorch Profiler trace
    parser.add_argument("--trace-file", type=str, default=None,
                        help="Output PyTorch Profiler trace file path")

    # optimizations - all are on by default but each can be disabled
    parser.add_argument("--disable_bf16", action="store_true",
                        help="Disables usage of torch.bfloat16")
    # torch.compile OR torch.export + AOTI OR neither
    parser.add_argument("--compile_export_mode", type=str, default="export_aoti",
                        choices=["compile", "export_aoti", "disabled"],
                        help="Configures how torch.compile or torch.export + AOTI are used")
    # fused (q, k, v) projections
    parser.add_argument("--disable_fused_projections", action="store_true",
                        help="Disables fused q,k,v projections")
    # channels_last memory format
    parser.add_argument("--disable_channels_last", action="store_true",
                        help="Disables usage of torch.channels_last memory format")
    # Flash Attention v3
    parser.add_argument("--disable_fa3", action="store_true",
                        help="Disables use of Flash Attention V3")
    # dynamic float8 quantization
    parser.add_argument("--disable_quant", action="store_true",
                        help="Disables usage of dynamic float8 quantization")
    # NSFW filter control
    parser.add_argument("--disable-nsfw-filter", action="store_true",
                        help="Disables NSFW content filter (allows all prompts)")
    # flags for tuning inductor
    parser.add_argument("--disable_inductor_tuning_flags", action="store_true",
                        help="Disables use of inductor tuning flags")
    return parser


# helper to annotate a function within a profiler trace
def annotate(f, title):
    @functools.wraps(f)
    def _f(*args, **kwargs):
        with record_function(title):
            return f(*args, **kwargs)
    return _f