Fix build errors and add Zero GPU support

- Add packages.txt for system dependencies (ffmpeg, mesa, OpenGL)
- Add spaces module import for Zero GPU decorator
- Fix Gradio API compatibility (remove show_share_button)
- Add PYOPENGL_PLATFORM environment variable
- Pin transformers==4.30.2 for PyTorch 2.1.0 compatibility
- Update Python version to 3.10.13 for Zero GPU support

README.md

@@ -1,15 +1,16 @@
 ---
 title: TANGO
-emoji: 🐠
+emoji: "🐠"
 colorFrom: blue
 colorTo: gray
 sdk: gradio
 sdk_version: 5.34.2
-python_version: 3.10.16
+python_version: 3.10.13
 app_file: app.py
 pinned: false
 short_description: Co-Speech Gesture Video Generation
 license: cc-by-nc-4.0
+suggested_hardware: zero-a10g
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,9 +1,12 @@
 import os
 import gradio as gr
 import shutil
-from inference import tango
 import numpy as np
 
+os.environ["PYOPENGL_PLATFORM"] = "osmesa"
+
+from inference import tango
+
 
 SCRIPT_PATH = os.path.dirname(os.path.realpath(__file__))
 
@@ -102,7 +105,6 @@ def make_demo():
                     interactive=False,
                     autoplay=False,
                     loop=False,
-                    show_share_button=True,
                 )
             with gr.Column(scale=4):
                 video_output_2 = gr.Video(
@@ -110,7 +112,6 @@ def make_demo():
                     interactive=False,
                     autoplay=False,
                     loop=False,
-                    show_share_button=True,
                 )
             with gr.Column(scale=1):
                 file_output_1 = gr.File(label="Download 3D Motion and Visualize in Blender")

inference.py

@@ -22,6 +22,13 @@ import smplx
 import igraph
 import fire
 
+try:
+    import spaces
+    ZERO_GPU_AVAILABLE = True
+except ImportError:
+    ZERO_GPU_AVAILABLE = False
+    spaces = None
+
 from utils.video_io import save_videos_from_pil
 from utils.genextend_inference_utils import adjust_statistics_to_match_reference
 from create_graph import path_visualization, graph_pruning, get_motion_reps_tensor, path_visualization_v2
@@ -526,7 +533,13 @@ TARGET_SR = 16000
 OUTPUT_DIR = os.path.join(SCRIPT_PATH, "outputs/")
 
 
-# @spaces.GPU(duration=200)
+def _gpu_decorator(func):
+    if ZERO_GPU_AVAILABLE and spaces is not None:
+        return spaces.GPU(duration=200)(func)
+    return func
+
+
+@_gpu_decorator
 def tango(audio_path, character_name, seed=2024, create_graph=False, video_folder_path=None):
     shutil.rmtree(OUTPUT_DIR, ignore_errors=True)
     os.makedirs(OUTPUT_DIR, exist_ok=True)

packages.txt

@@ -0,0 +1,8 @@
+ffmpeg
+libgl1-mesa-glx
+libglib2.0-0
+libsm6
+libxext6
+libxrender1
+libosmesa6-dev
+freeglut3-dev

requirements.txt

@@ -1,3 +1,5 @@
+spaces
+huggingface_hub
 gradio
 scikit-image==0.21.0
 scikit-learn==1.3.2
@@ -10,7 +12,7 @@ opencv-python==4.8.1.78
 tensorboardx
 filterpy
 cython
-Pillow==9.5.0
+Pillow>=9.5.0
 trimesh
 pyrender
 matplotlib
@@ -23,7 +25,7 @@ easydict
 pycocotools
 plyfile
 timm
-pyglet
+pyglet<2.0.0
 eval_type_backport
 wandb
 wget
@@ -31,8 +33,8 @@ av==11.0.0
 ffmpeg-python
 mediapipe
 batch-face @ git+https://github.com/elliottzheng/batch-face.git@master
-decord==0.6.0
-diffusers==0.24.0
+decord
+diffusers>=0.24.0
 imageio==2.33.0
 imageio-ffmpeg==0.4.9
 omegaconf==2.2.3
@@ -43,4 +45,5 @@ igraph
 ConfigArgParse
 librosa
 pytorch-fid
-fire
+fire
+soundfile