Initial commit

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+

pyproject.toml

@@ -0,0 +1,43 @@
+[build-system]
+requires = ["setuptools >= 75.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "flux-schnell-edge-inference"
+description = "An Optimization Pipeline"
+requires-python = ">=3.10,<3.13"
+version = "8"
+dependencies = [
+  "diffusers==0.31.0",
+  "transformers==4.46.2",
+  "accelerate==1.1.0",
+  "omegaconf==2.3.0",
+  "torch==2.5.1",
+  "protobuf==5.28.3",
+  "sentencepiece==0.2.0",
+  "torchao==0.6.1",
+  "optimum-quanto",
+  "hf_transfer==0.1.8",
+  "setuptools==75.2.0",
+  "edge-maxxing-pipelines @ git+https://github.com/womboai/edge-maxxing@7c760ac54f6052803dadb3ade8ebfc9679a94589#subdirectory=pipelines",
+]
+
+[[tool.edge-maxxing.models]]
+repository = "black-forest-labs/FLUX.1-schnell"
+revision = "741f7c3ce8b383c54771c7003378a50191e9efe9"
+exclude = ["transformer", "vae", "text_encoder_2"]
+
+[[tool.edge-maxxing.models]]
+repository = "city96/t5-v1_1-xxl-encoder-bf16"
+revision = "1b9c856aadb864af93c1dcdc226c2774fa67bc86"
+
+[[tool.edge-maxxing.models]]
+repository = "MyApricity/Vae_Only"
+revision = "a47d57702caf8ff0c0e21d30b93f9d3297b81920"
+
+[[tool.edge-maxxing.models]]
+repository = "MyApricity/Flux_Transformer_float8"
+revision = "66c5f182385555a00ec90272ab711bb6d3c197db"
+
+[project.scripts]
+start_inference = "main:main"

src/main.py

@@ -0,0 +1,50 @@
+from io import BytesIO
+from multiprocessing.connection import Listener
+from os import chmod, remove
+from os.path import abspath, exists
+from pathlib import Path
+
+from PIL.JpegImagePlugin import JpegImageFile
+from pipelines.models import TextToImageRequest
+
+from pipeline import load_pipeline, infer
+
+SOCKET = abspath(Path(__file__).parent.parent / "inferences.sock")
+
+
+def main():
+    print(f"Loading pipeline")
+    pipeline = load_pipeline()
+
+    print(f"Pipeline loaded! , creating socket at '{SOCKET}'")
+
+    if exists(SOCKET):
+        remove(SOCKET)
+
+    with Listener(SOCKET) as listener:
+        chmod(SOCKET, 0o777)
+
+        print(f"Awaiting connections")
+        with listener.accept() as connection:
+            print(f"Connected")
+
+            while True:
+                try:
+                    request = TextToImageRequest.model_validate_json(connection.recv_bytes().decode("utf-8"))
+                except EOFError:
+                    print(f"Inference socket exiting")
+
+                    return
+
+                image = infer(request, pipeline)
+
+                data = BytesIO()
+                image.save(data, format=JpegImageFile.format)
+
+                packet = data.getvalue()
+
+                connection.send_bytes(packet)
+
+
+if __name__ == '__main__':
+    main()

src/pipeline.py

@@ -0,0 +1,134 @@
+import os
+import torch
+import torch._dynamo
+import gc
+
+import json
+import transformers
+from huggingface_hub.constants import HF_HUB_CACHE
+from transformers import T5EncoderModel, T5TokenizerFast, CLIPTokenizer, CLIPTextModel
+
+from torchao.quantization import quantize_, int8_weight_only, fpx_weight_only
+from torch import Generator
+from diffusers import FluxTransformer2DModel, DiffusionPipeline
+
+from PIL.Image import Image
+from diffusers import FluxPipeline, AutoencoderKL, AutoencoderTiny
+from pipelines.models import TextToImageRequest
+from optimum.quanto import requantize
+import json
+
+
+
+
+torch._dynamo.config.suppress_errors = True
+os.environ['PYTORCH_CUDA_ALLOC_CONF']="expandable_segments:True"
+os.environ["TOKENIZERS_PARALLELISM"] = "True"
+
+CHECKPOINT = "black-forest-labs/FLUX.1-schnell"
+REVISION = "741f7c3ce8b383c54771c7003378a50191e9efe9"
+Pipeline = None
+
+
+import torch
+import math
+from typing import Dict, Any
+
+def remove_cache():
+    gc.collect()
+    torch.cuda.empty_cache()
+    torch.cuda.reset_max_memory_allocated()
+    torch.cuda.reset_peak_memory_stats()
+
+
+class InitModel:
+
+    @staticmethod
+    def load_text_encoder() -> T5EncoderModel:
+        print("Loading text encoder...")
+        text_encoder = T5EncoderModel.from_pretrained(
+            "city96/t5-v1_1-xxl-encoder-bf16",
+            revision="1b9c856aadb864af93c1dcdc226c2774fa67bc86",
+            torch_dtype=torch.bfloat16,
+        )
+        return text_encoder.to(memory_format=torch.channels_last)
+
+    @staticmethod
+    def load_vae() -> AutoencoderTiny:
+        print("Loading VAE model...")
+        vae = AutoencoderTiny.from_pretrained(
+            "XiangquiAI/FLUX_Vae_Model",
+            revision="103bcc03998f48ef311c100ee119f1b9942132ab",
+            torch_dtype=torch.bfloat16,
+        )
+        return vae
+
+    @staticmethod
+    def load_transformer(trans_path: str) -> FluxTransformer2DModel:
+        print("Loading transformer model...")
+        transformer = FluxTransformer2DModel.from_pretrained(
+            trans_path,
+            torch_dtype=torch.bfloat16,
+            use_safetensors=False,
+        )
+        return transformer.to(memory_format=torch.channels_last)
+
+
+
+def load_pipeline() -> Pipeline:
+
+
+    transformer_path = os.path.join(HF_HUB_CACHE, "models--MyApricity--Flux_Transformer_float8/snapshots/66c5f182385555a00ec90272ab711bb6d3c197db")
+    transformer = InitModel.load_transformer(transformer_path)
+    
+    text_encoder_2 = InitModel.load_text_encoder()
+    vae = InitModel.load_vae()
+
+
+    pipeline = DiffusionPipeline.from_pretrained(CHECKPOINT, 
+                        revision=REVISION, 
+                        vae=vae, 
+                        transformer=transformer, 
+                        text_encoder_2=text_encoder_2, 
+                        torch_dtype=torch.bfloat16)
+    pipeline.to("cuda")
+    try:
+        pipeline.disable_vae_slice()
+    except:
+        print("Using origin pipeline")
+        
+
+    promts_listing = [
+        "melanogen, endosome",
+        "buffer, cutie, buttinsky, prototrophic",
+        "puzzlehead, fistical, must return non duplicate",
+        "apical, polymyodous, tiptilt"
+    ]
+    
+    for p in promts_listing:
+        pipeline(prompt=p, 
+                        width=1024,
+                        height=1024,
+                        guidance_scale=0.0,
+                        num_inference_steps=4,
+                        max_sequence_length=256)
+
+    return pipeline
+
+
+@torch.no_grad()
+def infer(request: TextToImageRequest, pipeline: Pipeline) -> Image:
+
+    remove_cache()
+    # remove cache here for better result
+    generator = Generator(pipeline.device).manual_seed(request.seed)
+
+    return pipeline(
+        request.prompt,
+        generator=generator,
+        guidance_scale=0.0,
+        num_inference_steps=4,
+        max_sequence_length=256,
+        height=request.height,
+        width=request.width,
+    ).images[0]