llm配置不生效，使用了agent的大模型 #32
Description
通过llama.cpp 加载了Qwen3-Embedding-8B-Q4_K_M 和 Qwen3.5-9B-IQ4_XS，在配置文件里配置了embdding和llm,仅embdding生效，通过启动日志查看，llm调用的是agent的模型
openclaw配置如下：
"agents": {
"defaults": {
"model": {
"primary": "vllm/Qwen3.5-35B-A3B"
},
"models": {
"vllm/Qwen3.5-35B-A3B": {}
},
"workspace": "/home/mc/.openclaw/workspace",
"compaction": {
"mode": "safeguard"
}
}
}
"graph-memory": {
"enabled": true,
"config": {
"llm": {
"apiKey": "111111111111111111111111111",
"baseURL": "http://localhost:8080/v1",
"model": "Qwen3.5-9B-IQ4_XS"
},
"embedding": {
"apiKey": "1111111111111111111111111111",
"baseURL": "http://localhost:8080/v1",
"model": "Qwen3-Embedding-8B-Q4_K_M",
"dimensions": 512
}
}
}
Qwen3.5-9B-IQ4_XS测试能正常访问：
(base) mc@mc-X570-Taichi:~$ curl http://localhost:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen3.5-9B-IQ4_XS",
"messages": [
{"role": "system", "content": "你是一个助手。"},
{"role": "user", "content": "介绍一下自己"}
],
"temperature": 0.7,
"stream": false
}'
{"choices":[{"finish_reason":"stop","index":0,"message":{"role":"assistant","content":"你好！我是你的智能助手。你可以叫我助手。\n\n我致力于为你提供高效、准确且友好的帮助。无论你是在寻找信息、需要创意灵感、想要解决复杂的逻辑问题，还是仅仅想进行轻松的闲聊，我都会尽力协助你。\n\n**我可以为你做什么？**\n* **解答疑惑**：回答关于事实、知识、技术或生活常识的问题。\n* **内容创作**：帮你写邮件、文章、故事、诗歌或营销文案。\n* **逻辑与分析**：总结长篇文章、分析数据、梳理思路或提供决策建议。\n* **语言学习**：辅助翻译、纠正语法或模拟对话练习。\n* **编程辅助**：编写、解释或调试代码片段。\n\n虽然我很努力，但我也会犯错，如果有任何地方做得不够好，欢迎随时指出，我会努力改进！\n\n今天有什么我可以帮你的吗？"}}],"created":1774717098,"model":"Qwen3.5-9B-IQ4_XS","system_fingerprint":"b8533-0fac87b15","object":"chat.completion","usage":{"completion_tokens":185,"prompt_tokens":23,"total_tokens":208,"prompt_tokens_details":{"cached_tokens":0}},"id":"chatcmpl-XFQ5rS50TTRLpv3al8MTQYTlA7OIzeKZ","timings":{"cache_n":0,"prompt_n":23,"prompt_ms":110.298,"prompt_per_token_ms":4.795565217391305,"prompt_per_second":208.52599321837207,"predicted_n":185,"predicted_ms":2726.55,"predicted_per_token_ms":14.73810810810811,"predicted_per_second":67.85131393152518}}
启动日志：用的llm是Qwen3.5-35B-A3B
(base) mc@mc-X570-Taichi:~$ openclaw gateway --verbose
🦞 OpenClaw 2026.3.24 (cff6dc9) — I'm the middleware between your ambition and your attention span.
00:50:51+08:00 [gateway] [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: graph-memory (/home/mc/.openclaw/extensions/graph-memory/index.ts), openclaw-weixin (/home/mc/.openclaw/extensions/openclaw-weixin/index.ts). Set plugins.allow to explicit trusted ids.
00:50:51+08:00 [gateway] [graph-memory] ready | db=~/.openclaw/graph-memory.db | provider=vllm | model=Qwen3.5-35B-A3B
00:50:54+08:00 [canvas] host mounted at http://127.0.0.1:18789/__openclaw__/canvas/ (root /home/mc/.openclaw/canvas)
00:50:54+08:00 bonjour: starting (hostname=openclaw, instance="mc-X570-Taichi (OpenClaw)", gatewayPort=18789, minimal=true)
00:50:54+08:00 [heartbeat] started
00:50:54+08:00 [health-monitor] started (interval: 300s, startup-grace: 60s, channel-connect-grace: 120s)
00:50:54+08:00 [gateway] agent model: vllm/Qwen3.5-35B-A3B
00:50:54+08:00 [gateway] listening on ws://127.0.0.1:18789, ws://[::1]:18789 (PID 17416)
00:50:54+08:00 [gateway] log file: /tmp/openclaw/openclaw-2026-03-29.log
00:50:54+08:00 [gateway] [graph-memory] vector search ready
00:50:54+08:00 [browser/server] Browser control listening on http://127.0.0.1:18791/ (auth=token)
00:50:54+08:00 [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: graph-memory (/home/mc/.openclaw/extensions/graph-memory/index.ts), openclaw-weixin (/home/mc/.openclaw/extensions/openclaw-weixin/index.ts). Set plugins.allow to explicit trusted ids.
00:50:54+08:00 [plugins] [graph-memory] ready | db=~/.openclaw/graph-memory.db | provider=vllm | model=Qwen3.5-35B-A3B
00:50:54+08:00 [plugins] [graph-memory] ready | db=~/.openclaw/graph-memory.db | provider=vllm | model=Qwen3.5-35B-A3B
00:50:54+08:00 [plugins] [graph-memory] ready | db=~/.openclaw/graph-memory.db | provider=vllm | model=Qwen3.5-35B-A3B
00:50:55+08:00 [plugins] [graph-memory] ready | db=~/.openclaw/graph-memory.db | provider=vllm | model=Qwen3.5-35B-A3