cold-start fix: GPU onnx runtime + shutdown_after_seconds=86400

- chute_config: onnxruntime-gpu, shutdown 300->86400, timeout 900, min_vram 16, max_hourly_price 2
- miner.py: CUDA provider with CPU fallback

chute_config.yml

@@ -2,17 +2,20 @@ Image:
   from_base: parachutes/python:3.12
   run_command:
     - pip install --upgrade setuptools wheel
-    - pip install huggingface_hub==0.19.4 onnxruntime==1.* opencv-python-headless numpy pydantic
+    - pip install 'huggingface_hub>=0.19.4' 'onnxruntime-gpu[cuda,cudnn]>=1.16' 'opencv-python-headless>=4.7' 'numpy>=1.23' 'pydantic>=2.0' 'pyyaml>=6.0' 'aiohttp>=3.9'
   set_workdir: /app
 
 NodeSelector:
   gpu_count: 1
+  min_vram_gb_per_gpu: 16
+  max_hourly_price_per_gpu: 2
   include:
     - pro_6000
 
 Chute:
   tee: true
-  shutdown_after_seconds: 300
+  timeout_seconds: 900
+  shutdown_after_seconds: 86400
   concurrency: 4
   max_instances: 5
   scaling_threshold: 0.5

miner.py

@@ -52,11 +52,15 @@ class Miner:
         so.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
         self.sess = ort.InferenceSession(
             str(Path(path_hf_repo) / self.weights_file),
-            providers=["CPUExecutionProvider"],
+            providers=[
+                ("CUDAExecutionProvider", {"device_id": 0}),
+                "CPUExecutionProvider",
+            ],
             sess_options=so,
         )
         self.inp = self.sess.get_inputs()[0].name
-        print("✅ ONNX beverage model loaded")
+        active = self.sess.get_providers()[0]
+        print(f"✅ ONNX beverage model loaded (provider={active})")
 
     def __repr__(self) -> str:
         return f"BeverageONNX(in={self.input_size}, cls={self.num_classes})"