.

Dockerfile

@@ -27,8 +27,8 @@ EXPOSE 8001
 # 新增：给 PT 显存分段配置，减少碎片 (可选但推荐)
 ENV PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:32
 
-# 启动：Gunicorn + 4 worker，每个 worker 一个独立进程
+# 启动：Gunicorn + 1 worker，每个 worker 一个独立进程
 CMD ["gunicorn", "app.main:app", \
      "-k", "uvicorn.workers.UvicornWorker", \
-     "-w", "4", \
+     "-w", "1", \
      "-b", "0.0.0.0:8001"]

app/main.py

@@ -27,9 +27,9 @@ from FlagEmbedding import BGEM3FlagModel
 # Config
 # -----------------------------------------------------------------------------#
 MODEL_PATH = "model/bge-m3"          # 按需改成你的权重路径
-MODEL_VRAM_MB = int(os.getenv("MODEL_VRAM_MB", "16384"))   # bge-m3-large fp16 ≈ 16 GiB
-POST_LOAD_GAP_MB = 512     
-SAFE_MIN_FREE_MB = MODEL_VRAM_MB + POST_LOAD_GAP_MB      # == 16896 MB
+MODEL_VRAM_MB = int(os.getenv("MODEL_VRAM_MB", "8000"))   # bge-m3-large fp32 ≈ 8 GiB
+POST_LOAD_GAP_MB = 192     
+SAFE_MIN_FREE_MB = MODEL_VRAM_MB + POST_LOAD_GAP_MB      # == 8192 MB
 
 # -----------------------------------------------------------------------------#
 # Logging
@@ -179,12 +179,24 @@ 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():
-    logger.info("Warm-up on %s", DEVICE)
-    try:
-        _ = model.encode(["warmup"], return_dense=True, num_processes=1)
-    except Exception as e:
-        logger.warning("Warm-up failed: %s — 首条请求时再退避", e)
+def warm_up():
+    def _warm_worker(t, q):
+        try:
+            _ = model.encode(t, return_dense=True, num_processes=1)
+            q.put("ok")
+        except Exception as e:
+            q.put(str(e))
+
+    texts = ["warmup"]
+    q = mp.Queue()
+    p = mp.Process(target=_warm_worker, args=(texts, q))
+    p.start()
+    p.join(timeout=60)
+
+    if not q.empty() and q.get() == "ok":
+        logger.info("Warm-up complete.")
+    else:
+        logger.warning("Warm-up failed or timed out.")