src/pipeline.py

from diffusers import (
    DiffusionPipeline,
    AutoencoderKL,
    FluxPipeline,
    FluxTransformer2DModel
)
from diffusers.image_processor import VaeImageProcessor
from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
from huggingface_hub.constants import HF_HUB_CACHE
from transformers import (
    T5EncoderModel,
    T5TokenizerFast,
    CLIPTokenizer,
    CLIPTextModel
)
import torch
import torch._dynamo
import gc
from PIL import Image
from pipelines.models import TextToImageRequest
from torch import Generator
import time
import math
from typing import Type, Dict, Any, Tuple, Callable, Optional, Union
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
from torchao.quantization import quantize_, float8_weight_only, int4_weight_only
from para_attn.first_block_cache.diffusers_adapters import apply_cache_on_pipe as cacher
import os
os.environ['PYTORCH_CUDA_ALLOC_CONF']="expandable_segments:True"
os.environ["TOKENIZERS_PARALLELISM"] = "True"
torch._dynamo.config.suppress_errors = True
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
torch.cuda.set_per_process_memory_fraction(0.99)

# globals
Pipeline = None
ckpt_id = "manbeast3b/flux.1-schnell-full1"
ckpt_revision = "cb1b599b0d712b9aab2c4df3ad27b050a27ec146"

def load_pipeline() -> Pipeline:
    model_name = "manbeast3b/Flux.1.Schnell-full-quant1"
    revision = "e7ddf488a4ea8a3cba05db5b8d06e7e0feb826a2"

    hub_model_dir = os.path.join(
        HF_HUB_CACHE,
        f"models--{model_name.replace('/', '--')}",
        "snapshots",
        revision,
        "transformer"
    )
    transformer = FluxTransformer2DModel.from_pretrained(
        hub_model_dir,
        torch_dtype=torch.bfloat16,
        use_safetensors=False
    ).to(memory_format=torch.channels_last)
    
    pipeline = FluxPipeline.from_pretrained(
        ckpt_id,
        revision=ckpt_revision,
        # text_encoder_2=text_encoder_2,
        transformer=transformer,
        # vae=vae,
        torch_dtype=torch.bfloat16
    )
    # pipeline.vae = torch.compile(vae)
    pipeline.to("cuda")
    pipeline = cacher(pipeline,residual_diff_threshold=0.56)
    quantize_(pipeline.vae, int4_weight_only())
    
    warmup_ = "controllable varied focus thai warriors entertainment claude still goat gang gang yeah"
    for _ in range(1):
        pipeline(
            prompt=warmup_,
            width=1024,
            height=1024,
            guidance_scale=0.0,
            num_inference_steps=4,
            max_sequence_length=256
        )
    return pipeline

sample = 1
@torch.no_grad()
def infer(request: TextToImageRequest, pipeline: Pipeline, generator: Generator) -> Image:
    global sample
    if not sample:
        sample=1
        gc.collect()
        torch.cuda.empty_cache()
        torch.cuda.reset_max_memory_allocated()
        torch.cuda.reset_peak_memory_stats()
        
    return pipeline(request.prompt,generator=generator, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256, 
                    height=request.height, width=request.width, output_type="pil").images[0]