diff --git a/api/onnx_web/utils.py b/api/onnx_web/utils.py
index 54c913e8..4755c0ee 100644
--- a/api/onnx_web/utils.py
+++ b/api/onnx_web/utils.py
@@ -97,10 +97,12 @@ def run_gc(devices: Optional[List[DeviceParams]] = None):
                 torch.cuda.empty_cache()
                 torch.cuda.ipc_collect()
                 mem_free, mem_total = torch.cuda.mem_get_info()
+                mem_pct = (1 - (mem_free / mem_total)) * 100
                 logger.debug(
-                    "remaining CUDA VRAM usage: %s of %s",
+                    "CUDA VRAM usage: %s of %s (%.2f%%)",
                     (mem_total - mem_free),
                     mem_total,
+                    mem_pct,
                 )
 
 
diff --git a/api/onnx_web/worker/pool.py b/api/onnx_web/worker/pool.py
index d9b838da..60d0930f 100644
--- a/api/onnx_web/worker/pool.py
+++ b/api/onnx_web/worker/pool.py
@@ -291,7 +291,9 @@ class DevicePoolExecutor:
                 logger.debug("shutting down worker for device %s", device)
                 worker.join(self.join_timeout)
                 if worker.is_alive():
-                    logger.error("leaking worker for device %s could not be shut down", device)
+                    logger.error(
+                        "leaking worker for device %s could not be shut down", device
+                    )
 
             self.leaking[:] = [dw for dw in self.leaking if dw[1].is_alive()]