handler.py

import os
from typing import Any, Dict

from diffusers import FluxPipeline, FluxTransformer2DModel, AutoencoderKL, TorchAoConfig
from PIL import Image
import torch

IS_COMPILE = True

if IS_COMPILE:
    import torch._dynamo
    torch._dynamo.config.suppress_errors = True

#from huggingface_inference_toolkit.logging import logger

def compile_pipeline(pipe) -> Any:
    pipe.transformer.fuse_qkv_projections()
    pipe.transformer.to(memory_format=torch.channels_last)
    pipe.transformer = torch.compile(pipe.transformer, mode="reduce-overhead", fullgraph=False, dynamic=False, backend="inductor")
    return pipe

class EndpointHandler:
    def __init__(self, path=""):
        repo_id = "camenduru/FLUX.1-dev-diffusers"
        #repo_id = "NoMoreCopyright/FLUX.1-dev-test"
        dtype = torch.bfloat16
        quantization_config = TorchAoConfig("int8wo")
        vae = AutoencoderKL.from_pretrained(repo_id, subfolder="vae", torch_dtype=dtype)
        #transformer = FluxTransformer2DModel.from_pretrained(repo_id, subfolder="transformer", torch_dtype=dtype, quantization_config=quantization_config).to("cuda")
        self.pipeline = FluxPipeline.from_pretrained(repo_id, vae=vae, torch_dtype=dtype, quantization_config=quantization_config)
        if IS_COMPILE: self.pipeline = compile_pipeline(self.pipeline)
        self.pipeline.to("cuda")

    #@torch.inference_mode()
    def __call__(self, data: Dict[str, Any]) -> Image.Image:
        #logger.info(f"Received incoming request with {data=}")

        if "inputs" in data and isinstance(data["inputs"], str):
            prompt = data.pop("inputs")
        elif "prompt" in data and isinstance(data["prompt"], str):
            prompt = data.pop("prompt")
        else:
            raise ValueError(
                "Provided input body must contain either the key `inputs` or `prompt` with the"
                " prompt to use for the image generation, and it needs to be a non-empty string."
            )

        parameters = data.pop("parameters", {})

        num_inference_steps = parameters.get("num_inference_steps", 28)
        width = parameters.get("width", 1024)
        height = parameters.get("height", 1024)
        guidance_scale = parameters.get("guidance_scale", 3.5)

        # seed generator (seed cannot be provided as is but via a generator)
        seed = parameters.get("seed", 0)
        generator = torch.manual_seed(seed)

        return self.pipeline(  # type: ignore
            prompt,
            height=height,
            width=width,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            generator=generator,
        ).images[0]