fix: bake GGUF model into Docker image during build

Dockerfile

@@ -15,6 +15,13 @@ RUN curl -sL https://github.com/ggml-org/llama.cpp/releases/download/b9208/llama
     ldconfig && \
     rm /tmp/llama.tar.gz
 
+# 下载 GGUF 多模态模型（构建时下载，约 1.7GB）
+RUN mkdir -p /app/models && \
+    curl -sL -o /app/models/MiniCPM-V-4.6-Q4_K_M.gguf \
+    "https://huggingface.co/ggml-org/MiniCPM-V-4.6-GGUF/resolve/main/MiniCPM-V-4.6-Q4_K_M.gguf" && \
+    curl -sL -o /app/models/mmproj-MiniCPM-V-4.6-Q8_0.gguf \
+    "https://huggingface.co/ggml-org/MiniCPM-V-4.6-GGUF/resolve/main/mmproj-MiniCPM-V-4.6-Q8_0.gguf"
+
 COPY requirements.txt .
 RUN pip install -r requirements.txt --no-cache-dir
 

app.py

@@ -1,11 +1,10 @@
 """
 OpenWolf 视觉 Space — llama-server 二进制驱动（GGUF / MiniCPM-V / 多模态）
+模型已内置在 Docker 镜像中
 """
 import os, time, threading, subprocess, requests, uuid
-from pathlib import Path
 from fastapi import FastAPI, Request, HTTPException
 from fastapi.responses import JSONResponse
-from huggingface_hub import hf_hub_download
 
 app = FastAPI(title="OpenWolf Vision")
 
@@ -13,10 +12,8 @@ LLAMA_PORT = 8081
 _ready = False
 _llama_proc = None
 _tasks = {}
-
-MODEL_REPO = "ggml-org/MiniCPM-V-4.6-GGUF"
-MODEL_FILE = "MiniCPM-V-4.6-Q4_K_M.gguf"
-MMPROJ_FILE = "mmproj-MiniCPM-V-4.6-Q8_0.gguf"
+MODEL_PATH = "/app/models/MiniCPM-V-4.6-Q4_K_M.gguf"
+MMPROJ_PATH = "/app/models/mmproj-MiniCPM-V-4.6-Q8_0.gguf"
 
 
 @app.on_event("startup")
@@ -26,25 +23,16 @@ async def startup():
 
 def _startup():
     global _ready, _llama_proc
-    os.makedirs("/app/models", exist_ok=True)
 
-    print("[startup] 下载多模态模型...")
-    t0 = time.time()
-    try:
-        model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE, local_dir="/app/models")
-        mmproj_path = hf_hub_download(repo_id=MODEL_REPO, filename=MMPROJ_FILE, local_dir="/app/models")
-        print(f"[startup] 下载完成 ({time.time()-t0:.1f}s)")
-    except Exception as e:
-        print(f"[startup] 下载失败: {e}")
+    if not os.path.exists(MODEL_PATH):
+        print(f"[startup] 模型文件不存在: {MODEL_PATH}")
         return
 
     print("[startup] 启动 llama-server (多模态)...")
-    _llama_proc = subprocess.Popen([
-        "llama-server", "--model", model_path,
-        "--mmproj", mmproj_path,
-        "--host", "127.0.0.1", "--port", str(LLAMA_PORT),
-        "--n-gpu-layers", "0", "--threads", "2", "--ctx-size", "4096",
-    ], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+    args = ["llama-server", "--model", MODEL_PATH, "--mmproj", MMPROJ_PATH,
+            "--host", "127.0.0.1", "--port", str(LLAMA_PORT),
+            "--n-gpu-layers", "0", "--threads", "2", "--ctx-size", "4096"]
+    _llama_proc = subprocess.Popen(args, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
 
     for _ in range(60):
         time.sleep(2)