Unload models from VRAM after use to free GPU memory

- Add unloadModel() method to LLMJudge class
- CLI calls unloadModel() after judging completes
- Workflows unload gemma3:4b after inference tests
- Uses Ollama API with keep_alive:0 to trigger unload

.github/workflows/full-pipeline.yml

@@ -165,6 +165,13 @@ jobs:
             exit 1
           fi
 
+      - name: Unload test model from VRAM
+        if: always()
+        run: |
+          echo "Unloading gemma3:4b from VRAM..."
+          curl -s http://localhost:11434/api/generate -d '{"model":"gemma3:4b","keep_alive":0}' || true
+          echo "Model unloaded"
+
       - name: Upload inference results
         uses: actions/upload-artifact@v4
         if: always()

.github/workflows/inference.yml

@@ -111,6 +111,13 @@ jobs:
             exit 1
           fi
 
+      - name: Unload test model from VRAM
+        if: always()
+        run: |
+          echo "Unloading gemma3:4b from VRAM..."
+          curl -s http://localhost:11434/api/generate -d '{"model":"gemma3:4b","keep_alive":0}' || true
+          echo "Model unloaded"
+
       - name: Upload inference results
         uses: actions/upload-artifact@v4
         if: always()