.

app/main.py

@@ -86,16 +86,22 @@ def load_model(device: str):
     use_fp16  = precision == "fp16"
     logger.info("Loading BGEM3 on %s (%s)", device, precision)
 
-    if device == "cpu":                          # ← 仅 CPU 路径
-        os.environ["CUDA_VISIBLE_DEVICES"] = ""  # 彻底摘掉 CUDA
+    if device == "cpu":
+        # CPU 路径：彻底摘掉 CUDA
+        os.environ["CUDA_VISIBLE_DEVICES"] = ""
         torch.cuda.is_available = lambda: False
         torch.cuda.device_count = lambda: 0
+    else:
+        # GPU 路径：只暴露本 worker 选中的那张卡
+        # 例：device == "cuda:3" → 只让当前进程看到 GPU 3
+        idx = device.split(":")[1]
+        os.environ["CUDA_VISIBLE_DEVICES"] = idx
+        # device_count 现在返回 1，BGEM3 只会在这张卡上建 1 个子进程
+        torch.cuda.device_count = lambda: 1
+
     mdl = BGEM3FlagModel(MODEL_PATH, use_fp16=use_fp16, device=device)
 
-    # # 多 GPU 时才包 DataParallel
-    # if device.startswith("cuda") and torch.cuda.device_count() > 1:
-    #     mdl = torch.nn.DataParallel(mdl)
-
+    # 不再包 DataParallel；每个 worker 单卡即可
     return mdl, precision
 
 # -----------------------------------------------------------------------------#