Migrate to ZeroGPU + VisCy v0.5.0a1

- Remove Dockerfile/runtime.txt; switch to native Gradio SDK for ZeroGPU.
- README: python_version 3.12, hardware zero-a10g.
- requirements.txt: install viscy-data/models/transforms/utils and cytoland
from mehta-lab/VisCy monorepo, pinned to alpha tag v0.5.0a1
(modular-viscy-staging @ 393abdf). Add spaces for @spaces.GPU.
- app.py: import VSUNet from cytoland.engine (new monorepo location), load
checkpoint to CPU, decorate predict() with @spaces.GPU, pick device inside
the decorated call.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Dockerfile

@@ -1,30 +0,0 @@
-FROM python:3.11-slim
-
-WORKDIR /code
-
-# Install system dependencies
-RUN apt-get update && apt-get install -y \
-    git \
-    git-lfs \
-    ffmpeg \
-    libsm6 \
-    libxext6 \
-    cmake \
-    rsync \
-    libgl1-mesa-glx \
-    && rm -rf /var/lib/apt/lists/* \
-    && git lfs install
-
-# Copy requirements and install Python dependencies
-COPY requirements.txt .
-RUN pip install --no-cache-dir --upgrade pip && \
-    pip install --no-cache-dir -r requirements.txt
-
-# Copy the rest of the application
-COPY . .
-
-# Expose the port Gradio runs on
-EXPOSE 7860
-
-# Command to run the application
-CMD ["python", "app.py"] 

README.md

@@ -8,7 +8,8 @@ sdk_version: 5.27.1
 app_file: app.py
 pinned: true
 license: bsd-3-clause
-python_version: 3.11
+python_version: "3.12"
+hardware: zero-a10g
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
+import spaces
 import torch
-from viscy.translation.engine import VSUNet
+from cytoland.engine import VSUNet
 from huggingface_hub import hf_hub_download
 from numpy.typing import ArrayLike
 import numpy as np
@@ -14,26 +15,19 @@ class VSGradio:
     def __init__(self, model_config, model_ckpt_path):
         self.model_config = model_config
         self.model_ckpt_path = model_ckpt_path
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-        print(f"Using device: {self.device}")
         self.model = None
         self.load_model()
 
     def load_model(self):
-        try:
-            # Load the model checkpoint and move it to the correct device (GPU or CPU)
-            print(f"Loading model from checkpoint: {self.model_ckpt_path}")
-            self.model = VSUNet.load_from_checkpoint(
-                self.model_ckpt_path,
-                architecture="UNeXt2_2D",
-                model_config=self.model_config,
-            )
-            self.model.to(self.device)
-            self.model.eval()
-            print("Model loaded successfully and set to evaluation mode")
-        except Exception as e:
-            print(f"Error loading model: {e}")
-            raise
+        print(f"Loading model from checkpoint: {self.model_ckpt_path}")
+        self.model = VSUNet.load_from_checkpoint(
+            self.model_ckpt_path,
+            architecture="UNeXt2_2D",
+            model_config=self.model_config,
+            map_location="cpu",
+        )
+        self.model.eval()
+        print("Model loaded on CPU; will move to GPU on demand via @spaces.GPU.")
 
     def normalize_fov(self, input: ArrayLike):
         "Normalizing the fov with zero mean and unit variance"
@@ -52,12 +46,17 @@ class VSGradio:
         new_width = int(width * scale_factor)
         return resize(inp, (new_height, new_width), anti_aliasing=True)
 
+    @spaces.GPU
     def predict(self, inp, scaling_factor: float):
         try:
             if inp is None:
                 print("Error: Input image is None")
                 return None, None
 
+            device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+            print(f"Running inference on device: {device}")
+            self.model.to(device)
+
             # Normalize the input and convert to tensor
             inp = self.normalize_fov(inp)
             original_shape = inp.shape
@@ -68,7 +67,7 @@ class VSGradio:
 
             test_dict = dict(
                 index=None,
-                source=inp.unsqueeze(0).unsqueeze(0).unsqueeze(0).to(self.device),
+                source=inp.unsqueeze(0).unsqueeze(0).unsqueeze(0).to(device),
             )
 
             with torch.inference_mode():

requirements.txt

@@ -1,6 +1,12 @@
-# Requires Python 3.11 or higher
-git+https://github.com/mehta-lab/VisCy.git@v0.3.0
+# Requires Python 3.12 or higher
+# VisCy monorepo (mehta-lab/VisCy) pinned to alpha tag v0.5.0a1
+# (points at modular-viscy-staging @ 393abdf7c62a4b72abdfca49bf1a210f5d5e5ec9)
+viscy-data @ git+https://github.com/mehta-lab/VisCy.git@v0.5.0a1#subdirectory=packages/viscy-data
+viscy-models @ git+https://github.com/mehta-lab/VisCy.git@v0.5.0a1#subdirectory=packages/viscy-models
+viscy-transforms @ git+https://github.com/mehta-lab/VisCy.git@v0.5.0a1#subdirectory=packages/viscy-transforms
+viscy-utils @ git+https://github.com/mehta-lab/VisCy.git@v0.5.0a1#subdirectory=packages/viscy-utils
+cytoland @ git+https://github.com/mehta-lab/VisCy.git@v0.5.0a1#subdirectory=applications/cytoland
+spaces
 gradio==5.27.1
 scikit-image
 cmap
-pydantic==2.11.3

runtime.txt

@@ -1 +0,0 @@
-python-3.11