[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

app.py

@@ -6,6 +6,7 @@ This source code is licensed under the license found in the
 LICENSE file in the root directory of this source tree.
 """
 
+import spaces  # noqa: F401  (must be imported before torch / any CUDA-touching module)
 from tempfile import NamedTemporaryFile
 import argparse
 import torch
@@ -250,6 +251,16 @@ def load_melody_filepath(melody_filepath, title, assigned_model, topp, temperatu
     
     return  gr.update(value=melody_name), gr.update(maximum=MAX_PROMPT_INDEX, value=-1), gr.update(value=assigned_model, interactive=True), gr.update(value=topp), gr.update(value=temperature), gr.update(value=cfg_coef), gr.update(maximum=MAX_OVERLAP)
 
+def _gpu_duration(model=None, text=None, melody_filepath=None, duration=10, *args, **kwargs):
+    """Rough per-call GPU time estimate for the ZeroGPU scheduler, scaled with requested audio duration."""
+    try:
+        d = int(duration)
+    except (TypeError, ValueError):
+        d = 10
+    return max(30, min(d * 3 + 30, 300))
+
+
+@spaces.GPU(duration=_gpu_duration)
 def predict(model, text, melody_filepath = None, duration=10, dimension=2, topk=200, topp=0, temperature=1.0, cfg_coef=4.0, background = None, title="UnlimitedMusicGen", settings_font="./assets/arial.ttf", settings_font_color = "#c87f05", seed=-1, overlap=1, prompt_index = 0, include_title = True, include_settings = True, harmony_only = False, profile = gr.OAuthProfile, segment_length = 30, settings_font_size=28, settings_animate_waveform=False, video_orientation="Landscape", excerpt_duration=3.5, return_history_json=False, progress=gr.Progress(track_tqdm=True)):
     """
     Generate music and video based on the provided parameters and model.
@@ -621,6 +632,7 @@ def fix_path(path: str) -> str:
         return "./" + path[index:].replace("\\", "/")
     return path
 # Add this wrapper function above the gr.api definitions
+@spaces.GPU(duration=_gpu_duration)
 def predict_simple(model: str, text: str,  melody_filepath: str = None, duration: int = 10, dimension: int = 2, topk: int = 200, topp: float = 0.01, temperature: float = 1.0, cfg_coef: float = 4.0, background: str = "./assets/background.png", title: str = "UnlimitedMusicGen", settings_font: str = "./assets/arial.ttf", settings_font_color: str = "#c87f05", seed: int = -1, overlap: int = 1, prompt_index: int = -1, include_title: bool = True, include_settings: bool = True, harmony_only: bool = False, profile: str = "Satoshi Nakamoto", segment_length: int = 30, settings_font_size: int = 28, settings_animate_waveform: bool = False, video_orientation: str = "Landscape",  return_history_json: bool = False) -> tp.List[tp.Tuple[str, str, str]]:
     """
         Generate music and video based on the provided parameters and model.

requirements.txt

@@ -4,14 +4,13 @@ einops
 flashy>=0.0.1
 hydra-core>=1.1
 hydra_colorlog
-torch==2.6.0 --extra-index-url https://download.pytorch.org/whl/cu124
-torchaudio>=2.0.0,<2.6.2 --extra-index-url https://download.pytorch.org/whl/cu124
+torchaudio
 soundfile
 huggingface_hub
 hf_xet
 tqdm
 transformers==4.43.4  # need Encodec there.
-xformers>=0.0.23 --index-url https://download.pytorch.org/whl/cu124
+xformers
 demucs
 librosa==0.11.0
 soundfile
@@ -29,7 +28,7 @@ mutagen
 fastapi>=0.88.0
 pydantic
 typer
-torchvision>=0.21.0 --extra-index-url https://download.pytorch.org/whl/cu124
+torchvision
 #torchtext
 pesq
 pystoi