.

app/main.py

@@ -86,13 +86,10 @@ def load_model(device: str):
     use_fp16 = precision == "fp16"
     logger.info("Loading BGEM3 on %s (%s)", device, precision)
 
-    # ---------- 关键：若走 CPU，先把整个 CUDA 伪装掉 ----------
-    if device == "cpu":
-        os.environ["CUDA_VISIBLE_DEVICES"] = ""      # 后续所有 fork 也看不到 GPU
+    if device == "cpu":                          # ← 仅 CPU 路径
+        os.environ["CUDA_VISIBLE_DEVICES"] = ""  # 彻底摘掉 CUDA
         torch.cuda.is_available = lambda: False
         torch.cuda.device_count = lambda: 0
-    # ----------------------------------------------------------
-
     mdl = BGEM3FlagModel(MODEL_PATH, use_fp16=use_fp16, device=device)
 
     # 多 GPU 时才包 DataParallel
@@ -173,11 +170,12 @@ tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
 app = FastAPI()
 logger.info("Using SAFE_MIN_FREE_MB = %d MB", SAFE_MIN_FREE_MB)
 
+# ② -------- FastAPI 启动预热 --------
 @app.on_event("startup")
-async def warm_up_mp_pool():
+async def warm_up():
     logger.info("Warm-up on %s", DEVICE)
     try:
-        _ = model.encode(["warmup"], return_dense=True)   # 不传 num_processes
+        _ = model.encode(["warmup"], return_dense=True)   # GPU 建池，CPU 单进程
     except Exception as e:
         logger.warning("Warm-up failed: %s — 首条请求时再退避", e)
 
@@ -195,9 +193,10 @@ def _encode(texts: List[str]):
       2. 若子进程 OOM / CUDA Error → 同一次请求 fallback 到 CPU
     绝不改全局状态，其他并发请求不受影响
     """
+    # ③ -------- _encode() 里 worker 调用 --------
     def _worker(t, q):
         try:
-            out = model.encode(t, return_dense=True)   # GPU / CPU 路径都安全
+            out = model.encode(t, return_dense=True)   # GPU or CPU 均安全
             q.put(("ok", out))
         except Exception as e:
             q.put(("err", str(e)))
@@ -220,31 +219,6 @@ def _encode(texts: List[str]):
 
     raise RuntimeError("子进程异常退出，无返回")
 
-# fallback_done = False  # prevent endless downgrade loop
-
-# def _encode(texts: List[str]):
-#     """Encode with single downgrade to CPU on OOM / CUDA failure."""
-#     global model, DEVICE, PRECISION, fallback_done
-
-#     try:
-#         return model.encode(texts, return_dense=True)
-
-#     except RuntimeError as err:
-#         is_oom = "out of memory" in str(err).lower()
-#         is_cuda_fail = "cuda error" in str(err).lower() or "device-side assert" in str(
-#             err
-#         ).lower()
-
-#         if (is_oom or is_cuda_fail) and not fallback_done:
-#             logger.error("GPU failure (%s). Falling back to CPU…", err)
-#             fallback_done = True
-#             torch.cuda.empty_cache()
-#             DEVICE = "cpu"
-#             model, PRECISION = load_model(DEVICE)
-#             return model.encode(texts, return_dense=True)
-
-#         raise  # second failure → propagate
-
 
 @app.post("/v1/embeddings")
 def create_embedding(request: EmbeddingRequest):