[Admin maintenance] Support new ZeroGPU hardware

Thank you so much for having shared this Space with the community on this demo. We have upgraded the ZeroGPU infra-structure to run on modern blackwell architecture.
For that, we need to upgrade your demo to support that. This PR fixes your demo to work with the new architecture. As this is something we broke on our end, we may merge this PR autonomously. If this breaks unexpectedly or brings unintended consequences, feel free to revert, modify or otherwise. Any issues you can email apolinario@huggingface.co

README.md

@@ -4,7 +4,8 @@ emoji: 🪄
 colorFrom: yellow
 colorTo: purple
 sdk: gradio
-sdk_version: 4.44.1
+sdk_version: 5.49.1
+python_version: "3.12"
 app_file: app.py
 pinned: false
 license: apache-2.0
@@ -13,4 +14,4 @@ models:
 - ali-vilab/ACE_Plus
 ---
 
-An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).
+An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).

app.py

@@ -2,21 +2,32 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
 import threading
 import time
-import gradio as gr
-import numpy as np
-import spaces
-import torch
-from PIL import Image
 import glob
 import os, csv, sys
 import shlex
 import subprocess
 import importlib
+
+# Python 3.12 removed `importlib.find_loader`; scepter still calls it. Shim it
+# back via `importlib.util.find_spec` so the scepter import chain succeeds.
+if not hasattr(importlib, "find_loader"):
+    import importlib.util
+    def _find_loader(name, path=None):
+        try:
+            spec = importlib.util.find_spec(name)
+        except (ImportError, ValueError):
+            return None
+        return spec.loader if spec is not None else None
+    importlib.find_loader = _find_loader
+
 subprocess.run("rm -rf /data-nvme/zerogpu-offload/*", env={}, shell=True)
 subprocess.run(shlex.split('pip install scepter --no-deps'))
-subprocess.run(shlex.split('pip install numpy==1.26'))
-subprocess.run(shlex.split('pip install flash-attn --no-build-isolation'),
-               env=os.environ | {'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"})
+
+import gradio as gr
+import numpy as np
+import spaces
+import torch
+from PIL import Image
 from scepter.modules.transform.io import pillow_convert
 from scepter.modules.utils.config import Config
 from scepter.modules.utils.distribute import we

requirements.txt

@@ -2,10 +2,8 @@ huggingface_hub
 diffusers
 transformers==4.49.0
 gradio_imageslider
-torch==2.4.0
-xformers==0.0.27.post2
-torchvision
-gradio==4.44.1
+torch==2.10.0
+torchvision==0.25.0
 opencv-python==4.10.0.84
 albumentations
 beautifulsoup4
@@ -23,4 +21,5 @@ scikit-learn
 sentencepiece
 tiktoken
 torchsde
-transformers-stream-generator
+transformers-stream-generator
+https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.1/flash_attn-2.8.1+cu12torch2.10cxx11abiTRUE-cp312-cp312-linux_x86_64.whl