flux_example.py

import logging
import time
import torch
import torch.distributed
from xfuser import xFuserFluxPipeline, xFuserArgs
from xfuser.config import FlexibleArgumentParser
from xfuser.core.distributed import (
    get_world_group,
    get_data_parallel_rank,
    get_data_parallel_world_size,
    get_runtime_state,
    is_dp_last_group,
)

def benchmark_torch_function(iters, f, *args, **kwargs):
    f(*args, **kwargs)
    f(*args, **kwargs)
    torch.cuda.synchronize()
    start_event = torch.cuda.Event(enable_timing=True)
    end_event = torch.cuda.Event(enable_timing=True)
    start_event.record()
    for _ in range(iters):
        f(*args, **kwargs)
    end_event.record()
    torch.cuda.synchronize()
    # elapsed_time has a resolution of 0.5 microseconds:
    # but returns milliseconds, so we need to multiply it to increase resolution
    return start_event.elapsed_time(end_event) * 1000 / iters, f(*args, **kwargs)



def main():
    parser = FlexibleArgumentParser(description="xFuser Arguments")
    args = xFuserArgs.add_cli_args(parser).parse_args()
    engine_args = xFuserArgs.from_cli_args(args)
    engine_config, input_config = engine_args.create_config()
    local_rank = get_world_group().local_rank

    pipe = xFuserFluxPipeline.from_pretrained(
        pretrained_model_name_or_path=engine_config.model_config.model,
        engine_config=engine_config,
        torch_dtype=torch.bfloat16,
    )

    if args.enable_sequential_cpu_offload:
        pipe.enable_sequential_cpu_offload(gpu_id=local_rank)
        logging.info(f"rank {local_rank} sequential CPU offload enabled")
    else:
        pipe = pipe.to(f"cuda:{local_rank}")

    pipe.prepare_run(input_config, max_sequence_length=256)

    torch.cuda.reset_peak_memory_stats()
    # start_time = time.time()
    avg_et, output = benchmark_torch_function(
        10,
        pipe,
        height=input_config.height,
        width=input_config.height,
        prompt=input_config.prompt,
        num_inference_steps=input_config.num_inference_steps,
        output_type=input_config.output_type,
        max_sequence_length=256,
        guidance_scale=0.0,
        generator=torch.Generator(device="cuda").manual_seed(input_config.seed),
    )
    # end_time = time.time()
    # elapsed_time = end_time - start_time
    peak_memory = torch.cuda.max_memory_allocated(device=f"cuda:{local_rank}")

    parallel_info = (
        f"dp{engine_args.data_parallel_degree}_cfg{engine_config.parallel_config.cfg_degree}_"
        f"ulysses{engine_args.ulysses_degree}_ring{engine_args.ring_degree}_"
        f"tp{engine_args.tensor_parallel_degree}_"
        f"pp{engine_args.pipefusion_parallel_degree}_patch{engine_args.num_pipeline_patch}"
    )
    if input_config.output_type == "pil":
        global_rank = get_world_group().rank
        dp_group_world_size = get_data_parallel_world_size()
        dp_group_index = global_rank // dp_group_world_size
        num_dp_groups = engine_config.parallel_config.dp_degree
        dp_batch_size = (input_config.batch_size + num_dp_groups - 1) // num_dp_groups
        if is_dp_last_group():
            for i, image in enumerate(output.images):
                image_rank = dp_group_index * dp_batch_size + i
                image.save(f"./results/flux_result_{parallel_info}_{image_rank}.png")
                print(
                    f"image {i} saved to ./results/flux_result_{parallel_info}_{image_rank}.png"
                )

    if get_world_group().rank == get_world_group().world_size - 1:
        print(f"epoch time: {avg_et / 1e6:.2f} sec, memory: {peak_memory/1e9} GB")
    get_runtime_state().destory_distributed_env()


if __name__ == "__main__":
    main()