Initial commit: LocalPilot：本地模型运行时，复用 llama-server 并提供 Ollama 兼容 provider

2026-09-19 11:57:59 +08:00
commit d159212416
27 changed files with 3022 additions and 0 deletions
@@ -0,0 +1,56 @@
+{
+  "runtime": {
+    "llama_server": "E:\\path\\to\\llama-server.exe",
+    "ollama_url": "http://127.0.0.1:11434",
+    "host": "127.0.0.1",
+    "port": 8787,
+    "default_model": "my-model",
+    "runtime_dir": "E:\\llm-backend\\runtime",
+    "cache_dir": "E:\\llm-backend\\cache",
+    "n_gpu_layers": "auto",
+    "ctx_size": 32768,
+    "flash_attn": "auto",
+    "fit_vram": true,
+    "auto_load": false,
+    "max_loaded_models": 1,
+    "keep_alive_seconds": 300,
+    "api_key": null
+  },
+  "models": [
+    {
+      "id": "my-gguf",
+      "kind": "gguf",
+      "path": "E:\\model\\my-model.Q4_K_M.gguf",
+      "enabled": true,
+      "options": {}
+    },
+    {
+      "id": "my-safetensors",
+      "kind": "transformers",
+      "path": "E:\\model\\my-model",
+      "enabled": true,
+      "options": {
+        "torch_dtype": "float16",
+        "device": "cuda",
+        "device_map": "auto",
+        "quantization": "none",
+        "load_in_4bit": false,
+        "dynamic_int8_cpu": false
+      }
+    },
+    {
+      "id": "my-ollama-model",
+      "kind": "ollama",
+      "remote_model": "mistral-nemo:latest",
+      "enabled": false
+    }
+  ],
+  "cloud_profiles": [
+    {
+      "id": "openai-compatible",
+      "base_url": "https://api.openai.com/v1",
+      "api_key_env": "OPENAI_API_KEY",
+      "default_model": "gpt-4o-mini"
+    }
+  ]
+}
@@ -0,0 +1,78 @@
+{
+  "runtime": {
+    "llama_server": "E:\\ollama\\模型\\llama-b8642-bin-win-cuda-13.1-x64\\llama-server.exe",
+    "ollama_url": "http://127.0.0.1:11434",
+    "host": "127.0.0.1",
+    "port": 8787,
+    "default_model": "apex-qwen3",
+    "model_dir": "E:\\model",
+    "runtime_dir": "E:\\llm-backend\\runtime",
+    "cache_dir": "E:\\llm-backend\\cache",
+    "n_gpu_layers": "auto",
+    "ctx_size": 32768,
+    "batch_size": 2048,
+    "ubatch_size": 512,
+    "flash_attn": "auto",
+    "fit_vram": true,
+    "auto_load": false,
+    "api_key": null,
+    "max_loaded_models": 1,
+    "keep_alive_seconds": 300
+  },
+  "models": [
+    {
+      "id": "apex-qwen3",
+      "kind": "transformers",
+      "path": "E:\\ollama\\Apex_AI-catmodel",
+      "enabled": true,
+      "description": "本机 Qwen3 safetensors 模型",
+      "options": {
+        "torch_dtype": "float16",
+        "device": "cuda",
+        "device_map": "auto",
+        "load_in_4bit": false,
+        "attn_implementation": "sdpa"
+      }
+    },
+    {
+      "id": "gguf-local",
+      "kind": "gguf",
+      "path": "E:\\model\\example.gguf",
+      "enabled": false,
+      "description": "将 path 改成真实 GGUF 文件后启用",
+      "options": {}
+    },
+    {
+      "id": "onnx-local",
+      "kind": "onnx",
+      "path": "E:\\model\\example.onnx",
+      "enabled": false,
+      "description": "ONNX Runtime/Optimum 扩展入口",
+      "options": {}
+    },
+    {
+      "id": "mistral-nemo-ollama",
+      "kind": "ollama",
+      "remote_model": "mistral-nemo-12b:latest",
+      "enabled": false,
+      "description": "兼容本机 Ollama 模型仓库；不复制权重"
+    },
+    {
+      "id": "deepseek-cloud",
+      "kind": "cloud",
+      "cloud_profile": "deepseek",
+      "remote_model": "deepseek-chat",
+      "enabled": false,
+      "description": "OpenAI 兼容云端模型"
+    }
+  ],
+  "cloud_profiles": [
+    {
+      "id": "deepseek",
+      "base_url": "https://api.deepseek.com",
+      "api_key_env": "DEEPSEEK_API_KEY",
+      "default_model": "deepseek-chat",
+      "timeout": 120
+    }
+  ]
+}