Add environment artifact: smoke_test.py

smoke_test.py

@@ -0,0 +1,85 @@
+"""Dependency, headless rendering and optional GPU checks; no model downloads."""
+import argparse
+import importlib
+import importlib.metadata
+import json
+import os
+import subprocess
+import tempfile
+from pathlib import Path
+
+os.environ.setdefault("PYOPENGL_PLATFORM", "osmesa")
+os.environ.setdefault("HF_HUB_OFFLINE", "1")
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--gpu", action="store_true")
+    args = parser.parse_args()
+    modules = [
+        "torch", "torchaudio", "torchvision", "numpy", "scipy", "soundfile",
+        "datasets", "transformers", "huggingface_hub", "safetensors", "yaml",
+        "f5_tts.api", "vocos", "smplx", "chumpy", "pyrender", "trimesh",
+        "imageio", "imageio_ffmpeg", "whisper", "jiwer", "python_speech_features",
+    ]
+    for name in modules:
+        importlib.import_module(name)
+    import numpy as np
+    import pyrender
+    import soundfile as sf
+    import torch
+    import torchaudio.functional as audio_fn
+    import trimesh
+
+    scene = pyrender.Scene(bg_color=[255, 255, 255, 255], ambient_light=[0.5, 0.5, 0.5])
+    material = pyrender.MetallicRoughnessMaterial(baseColorFactor=[0.55, 0.68, 0.82, 1.0])
+    scene.add(pyrender.Mesh.from_trimesh(trimesh.creation.icosphere(subdivisions=1), material=material))
+    pose = np.eye(4)
+    pose[2, 3] = 3.0
+    scene.add(pyrender.PerspectiveCamera(yfov=np.pi / 3), pose=pose)
+    scene.add(pyrender.DirectionalLight(color=np.ones(3), intensity=2), pose=pose)
+    renderer = pyrender.OffscreenRenderer(64, 64)
+    try:
+        color, depth = renderer.render(scene)
+        assert color.shape[:2] == (64, 64) and np.count_nonzero(depth) > 0
+    finally:
+        renderer.delete()
+    with tempfile.TemporaryDirectory() as temp:
+        wav = Path(temp) / "tone.wav"
+        sf.write(wav, np.zeros(24000, np.float32), 24000)
+        waveform, sr = sf.read(wav)
+        assert sr == 24000 and len(waveform) == 24000
+        video = Path(temp) / "test.mp4"
+        subprocess.run([
+            "ffmpeg", "-nostdin", "-v", "error", "-f", "lavfi", "-i",
+            "color=c=blue:s=64x64:r=30:d=1", "-i", str(wav),
+            "-c:v", "libx264", "-pix_fmt", "yuv420p", "-c:a", "aac",
+            "-t", "1", str(video),
+        ], check=True)
+        subprocess.run(["ffmpeg", "-nostdin", "-v", "error", "-xerror", "-i", str(video), "-f", "null", "-"], check=True)
+    tools = {}
+    for name, command in {
+        "git": ["git", "--version"], "git-lfs": ["git", "lfs", "version"],
+        "font": ["fc-match", "-f", "%{file}", "sans"],
+    }.items():
+        tools[name] = subprocess.check_output(command, text=True).strip()
+    assert Path(tools["font"]).is_file()
+    result = {
+        "tools": tools,
+        "imports": modules, "osmesa": "passed", "h264_aac": "passed",
+        "torch": torch.__version__, "cuda_build": torch.version.cuda,
+        "gpu_test_requested": args.gpu,
+    }
+    if args.gpu:
+        assert torch.cuda.is_available(), "A CUDA-capable runtime is required for --gpu"
+        x = torch.randn(128, 128, device="cuda")
+        assert torch.isfinite(x @ x.T).all().item()
+        resampled = audio_fn.resample(torch.zeros(1, 2400, device="cuda"), 24000, 16000)
+        assert resampled.shape[-1] == 1600
+        torch.cuda.synchronize()
+        result["gpu"] = torch.cuda.get_device_name(0)
+    print(json.dumps(result, ensure_ascii=False), flush=True)
+
+
+if __name__ == "__main__":
+    main()