[Admin maintenance] Support new ZeroGPU hardware

README.md

@@ -4,7 +4,8 @@ emoji: 🚀
 colorFrom: blue
 colorTo: indigo
 sdk: gradio
-sdk_version: 5.31.0
+sdk_version: 5.49.1
+python_version: "3.12"
 app_file: app.py
 pinned: false
 license: apache-2.0

app.py

@@ -1,15 +1,20 @@
+import os
+import ctypes
+
+# Preload CUDA 13 runtime so flash-attn's prebuilt cu13 wheel can find libcudart.so.13
+_CUDA_LIBDIR = "/cuda-image/usr/local/cuda-13.0/lib64"
+if os.path.isdir(_CUDA_LIBDIR):
+    os.environ["LD_LIBRARY_PATH"] = _CUDA_LIBDIR + os.pathsep + os.environ.get("LD_LIBRARY_PATH", "")
+    try:
+        ctypes.CDLL(os.path.join(_CUDA_LIBDIR, "libcudart.so.13"), mode=ctypes.RTLD_GLOBAL)
+    except OSError:
+        pass
+
 import spaces
 import gradio as gr
 import numpy as np
-import os
 import torch
 import random
-import subprocess
-subprocess.run(
-    "pip install flash-attn --no-build-isolation",
-    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-    shell=True,
-)
 
 from accelerate import infer_auto_device_map, load_checkpoint_and_dispatch, init_empty_weights
 from PIL import Image

requirements.txt

@@ -1,17 +1,18 @@
 decord==0.6.0
 einops==0.8.1
-huggingface_hub==0.29.1
-matplotlib==3.7.0
-numpy==1.24.4
-opencv_python==4.7.0.72
-pyarrow==11.0.0
-PyYAML==6.0.2
-Requests==2.32.3
-safetensors==0.4.5
-scipy==1.10.1
-sentencepiece==0.1.99
-torch==2.5.1
-torchvision==0.20.1
+huggingface_hub
+matplotlib
+numpy
+opencv_python
+pyarrow
+PyYAML
+Requests
+safetensors
+scipy
+sentencepiece
+torch==2.10.0
+torchvision==0.25.0
 transformers==4.49.0
 accelerate>=0.34.0
-wandb
+wandb
+https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.1/flash_attn-2.8.1+cu13torch2.10cxx11abiTRUE-cp312-cp312-linux_x86_64.whl