switch to Qwen3-VL-8B-Instruct, fix start.sh arg, enable tool calling with hermes parser

server.py: switch from os.execvp(vllm) to make_arg_parser + asyncio.run(run_server(args)) for better error handling; add GPU selection via CLI argument
opencode.json: add default model field and permission.question=ask to prevent auto-triggering question tool

opencode.json

@@ -1,5 +1,6 @@
 {
   "$schema": "https://opencode.ai/config.json",
+  "model": "Agent/Qwen3-VL-8B-Instruct",
   "provider": {
     "Agent": {
       "api": "openai",
@@ -8,15 +9,18 @@
         "apiKey": "not-needed"
       },
       "models": {
-        "Qwen2.5-7B-Instruct": {
-          "id": "Qwen2.5-7B-Instruct",
-          "name": "Qwen2.5-7B-Instruct",
+        "Qwen3-VL-8B-Instruct": {
+          "id": "Qwen3-VL-8B-Instruct",
+          "name": "Qwen3-VL-8B-Instruct",
           "limit": { "context": 32768, "output": 8192 },
           "cost": { "input": 0, "output": 0 }
         }
       }
     }
   },
+  "permission": {
+    "question": "ask"
+  },
   "experimental": {
     "mcp_timeout": 60000
   }

server.py

@@ -1,10 +1,18 @@
-import os, subprocess
+import os, sys
+import asyncio
+import subprocess
 from pathlib import Path
 
+from vllm.utils.argparse_utils import FlexibleArgumentParser
+from vllm.entrypoints.openai.api_server import run_server
+from vllm.entrypoints.openai.cli_args import make_arg_parser
+
+
 def _pick_free_gpu():
     if "CUDA_VISIBLE_DEVICES" in os.environ:
         return os.environ["CUDA_VISIBLE_DEVICES"]
     try:
+
         out = subprocess.check_output(
             ["nvidia-smi", "--query-gpu=index,memory.free", "--format=csv,noheader"],
             text=True,
@@ -14,27 +22,33 @@ def _pick_free_gpu():
         gpu = best.split(",")[0].strip()
     except Exception:
         gpu = "0"
-    os.environ["CUDA_VISIBLE_DEVICES"] = gpu
+        
     return gpu
 
+
 if __name__ == "__main__":
-    _pick_free_gpu()
-
-    model = str(Path(__file__).resolve().parent / "models/Qwen2.5-7B-Instruct")
+    gpu = _pick_free_gpu()
+    if len(sys.argv) > 1:
+        gpu = sys.argv[1]
 
+    os.environ["CUDA_VISIBLE_DEVICES"] = gpu
     os.environ["FLASHINFER_DISABLE_VERSION_CHECK"] = "1"
 
-    os.execvp("vllm", [
-        "vllm", "serve",
-        model,
+    model = str(Path(__file__).resolve().parent / "models/Qwen3-VL-8B-Instruct")
+
+    parser = make_arg_parser(FlexibleArgumentParser())
+    args = parser.parse_args([
+        "--model", model,
         "--host", "0.0.0.0",
         "--port", "8000",
-        "--served-model-name", "Qwen2.5-7B-Instruct",
+        "--served-model-name", "Qwen3-VL-8B-Instruct",
         "--trust-remote-code",
         "--max-model-len", "32768",
         "--dtype", "bfloat16",
-        "--gpu-memory-utilization", "0.85",
+        "--gpu-memory-utilization", "0.70",
         "--enforce-eager",
         "--enable-auto-tool-choice",
         "--tool-call-parser", "hermes",
     ])
+
+    asyncio.run(run_server(args))