siuuuuuuuuuu

Dockerfile

@@ -1,36 +1,30 @@
-# Use an official Python runtime as a parent image. 
-# 3.10-slim is lightweight but maintains high compatibility with ML libraries like PyTorch.
 FROM python:3.10-slim
 
-# Set environment variables to prevent Python from writing .pyc files 
-# and to ensure output is logged directly to the terminal.
 ENV PYTHONDONTWRITEBYTECODE=1
 ENV PYTHONUNBUFFERED=1
 ENV FLASK_APP=call_models.py
 
-# Set the working directory in the container
+# HuggingFace Spaces runs containers as uid=1000
+RUN useradd -m -u 1000 user
+
 WORKDIR /app
 
-# Install system dependencies
-# gcc and python3-dev are often required to compile certain Python C-extensions
 RUN apt-get update && apt-get install -y --no-install-recommends \
     gcc \
     python3-dev \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy only the requirements file first to leverage Docker cache for dependency installation
 COPY requirements.txt .
 
-# Install dependencies
-RUN pip install --no-cache-dir -r requirements.txt
+# Install CPU-only torch first to avoid pulling the ~2GB CUDA build,
+# then install the rest of the requirements.
+RUN pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu && \
+    pip install --no-cache-dir -r requirements.txt
+
+COPY --chown=user:user . .
 
-# Copy the rest of the application code, including the custom modules and models
-COPY . .
+USER user
 
-# Expose the port that the Flask app uses
 EXPOSE 7860
 
-# While your script uses `app.run()`, running a production WSGI server like Gunicorn 
-# is standard practice for handling concurrent requests and better resource management.
-# You can revert to `CMD ["python", "call_models.py"]` if you strictly want the built-in server.
-CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--workers", "2", "--timeout", "120", "call_models:app"]
+CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--workers", "2", "--timeout", "120", "call_models:app"]

call_models.py

@@ -94,8 +94,8 @@ def schedule():
     except ValueError:
         return jsonify({"error": f"invalid timestamp: {timestamp_str}"}), 400
 
-    ppo_path   = f"scheduler_{printer_id}_ppo"
-    model_path = f"model_{printer_id}.npz"
+    ppo_path   = f"scheduler_ppo"
+    model_path = f"model.npz"
 
     if not os.path.exists(f"{ppo_path}.zip"):
         return jsonify({"error": f"no PPO model found for printer {printer_id}"}), 404
@@ -118,4 +118,4 @@ def schedule():
 
 
 if __name__ == "__main__":
-    app.run(host="0.0.0.0", port=7860, debug=False)
+    app.run(host="0.0.0.0", port=7861, debug=False)

model.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d226c2b94e8ccd47fec4156e87a618a3421563b99d3c16c297b81851b81228d
+size 3506

predict_from_supabase.py

@@ -15,7 +15,13 @@ from scheduling_rl import _ACTION_TABLE, COMPONENT_NAMES
 # Supabase client
 # ---------------------------------------------------------------------------
 
-_sb = create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_SERVICE_KEY"])
+_sb = None
+
+def _get_sb():
+    global _sb
+    if _sb is None:
+        _sb = create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_SERVICE_KEY"])
+    return _sb
 
 # Column order must match the obs vector expected by the PPO (C=7 conditions)
 _CONDITION_COLS = [
@@ -46,8 +52,9 @@ _HEALTH_COLS = [
 
 def _fetch_health(printer_id: str, t: datetime) -> np.ndarray:
     """Snapshot whose time_step_id matches hours elapsed since last_repair."""
+    sb = _get_sb()
     printer = (
-        _sb.table("printers")
+        sb.table("printers")
         .select("last_repair")
         .eq("id", printer_id)
         .single()
@@ -61,7 +68,7 @@ def _fetch_health(printer_id: str, t: datetime) -> np.ndarray:
     time_step_id = int((t - last_repair).total_seconds() // 3600)
 
     row = (
-        _sb.table("snapshots")
+        sb.table("snapshots")
         .select(", ".join(_HEALTH_COLS))
         .eq("id", printer_id)
         .gte("time_step_id", time_step_id)
@@ -78,7 +85,7 @@ def _fetch_health(printer_id: str, t: datetime) -> np.ndarray:
 def _fetch_conditions(printer_id: str, t: datetime) -> np.ndarray:
     """Closest conditions row at or before t."""
     row = (
-        _sb.table("conditions")
+        _get_sb().table("conditions")
         .select(", ".join(_CONDITION_COLS))
         .eq("id", printer_id)
         .lte("timestamp", t.isoformat())

requirements.txt

@@ -12,6 +12,7 @@ supabase>=2.3.0
 
 # Required for loading PPO .zip models (Standard for Stable Baselines3)
 stable-baselines3>=2.2.1
+gymnasium>=0.29.0
 torch>=2.2.0  # Required backend for Stable Baselines3
 
 # Production WSGI Server (Recommended over standard Flask app.run)

scheduler_ppo.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8503dac424ae9073edfd74be9661097dc39b5319fd44b7fb16a40f9db2d603eb
+size 564738