Update handler.py

handler.py

@@ -1,5 +1,3 @@
-# https://github.com/sayakpaul/diffusers-torchao
-#6.22s
 import os
 from typing import Any, Dict
 from PIL import Image
@@ -7,26 +5,71 @@ import torch
 from diffusers import FluxPipeline
 from huggingface_inference_toolkit.logging import logger
 from para_attn.first_block_cache.diffusers_adapters import apply_cache_on_pipe
-# from torchao.quantization import quantize_, float8_dynamic_activation_float8_weight, float8_weight_only
 import time
+import uuid
+from huggingface_hub import HfApi
+
+from pyngrok import ngrok
+import subprocess
+from fastapi import FastAPI
+from fastapi.responses import FileResponse
+import uvicorn
+# Flask
+image_directory='./images'
+if not os.path.exists(image_directory):
+    os.makedirs(image_directory)
+    
+app = FastAPI()
+
+@app.get("/images/{image_name}")
+async def get_image(image_name: str):
+    image_path = os.path.join(image_directory, image_name)
+    
+    if os.path.exists(image_path):
+        return FileResponse(image_path)
+    else:
+        return {"error": "Image not found"}
+
+authtoken = "2cvqFKWc1eb9b0aN7pRLDUBfEtC_2FUehxFL8CAKXRkW3Hfjo"
+commands = [
+    "curl -sSL https://ngrok-agent.s3.amazonaws.com/ngrok.asc | sudo tee /etc/apt/trusted.gpg.d/ngrok.asc >/dev/null",
+    'echo "deb https://ngrok-agent.s3.amazonaws.com buster main" | sudo tee /etc/apt/sources.list.d/ngrok.list',
+    "sudo apt update",
+    "sudo apt install -y ngrok",
+    f"ngrok config add-authtoken {authtoken}",
+]
+for command in commands:
+    try:
+        subprocess.run(command, shell=True, check=True)
+        logger.info(f"SUCCESS CMD: {command}")
+    except subprocess.CalledProcessError as e:
+        logger.info(f"Failed CMD: {e}")
 
 class EndpointHandler:
-    def __init__(self,path=""):
+    def __init__(self, path=""):
         self.pipe = FluxPipeline.from_pretrained(
             "NoMoreCopyrightOrg/flux-dev",
             torch_dtype=torch.bfloat16,
         ).to("cuda")
         apply_cache_on_pipe(self.pipe, residual_diff_threshold=0.12)
-        # quantize_(self.pipe.text_encoder, float8_weight_only())
-        # quantize_(self.pipe.transformer, float8_dynamic_activation_float8_weight())
         self.pipe.transformer = torch.compile(
             self.pipe.transformer, mode="max-autotune-no-cudagraphs",
         )
         self.pipe.vae = torch.compile(
             self.pipe.vae, mode="max-autotune-no-cudagraphs",
         )
-
-    def __call__(self, data: Dict[str, Any]) -> Image.Image:
+        # Fastapi Run
+        uvicorn.run(app, host="127.0.0.1", port=5000)
+        # ngrok
+        self.public_url = ngrok.connect(5000).public_url
+        command = "ngrok http 5000"
+        try:
+            subprocess.run(command, shell=True, check=True)
+            print("ngrok HTTP run sucessfully")
+        except subprocess.CalledProcessError as e:
+            print(f"Falied ngrok: {e}")
+        
+    def __call__(self, data: Dict[str, Any]) -> str:
         logger.info(f"Received incoming request with {data=}")
 
         if "inputs" in data and isinstance(data["inputs"], str):
@@ -57,9 +100,14 @@ class EndpointHandler:
             guidance_scale=guidance_scale,
             num_inference_steps=num_inference_steps,
             generator=generator,
-            # output_type="pil" if dist.get_rank() == 0 else "pt",
         ).images[0]
         end_time = time.time()
         time_taken = end_time - start_time
         print(f"Time taken: {time_taken:.2f} seconds")
-        return result
+        filename = f"{uuid.uuid4()}.png"
+        image_path = f"/images/{filename}"
+
+        result.save(image_path)
+        image_url = f"{self.public_url+image_path}"
+
+        return image_url