Introduce Apple Silicon hardware optimization and dynamic LLM configuration

- Add `llm_config.py` to manage LLM model parameters dynamically via Apple Silicon detection.
- Implement `apple_silicon.py` utility to retrieve and summarize hardware specifications.
- Refactor `generator.py` to leverage dynamic parameter fetching and optimize threading for performance cores.
- Enhance `system.py` with `is_apple_silicon` check and augment system info output.

src/voice_dialogue/config/llm_config.py

@@ -0,0 +1,110 @@
+"""LLM模型配置管理"""
+
+from typing import Dict, Any
+from voice_dialogue.utils.apple_silicon import get_optimal_llama_cpp_config, get_apple_silicon_info
+
+__all__ = ('get_llm_model_params', 'get_apple_silicon_summary')
+
+
+def get_llm_model_params() -> Dict[str, Any]:
+    """
+    获取LLM模型参数，基于Apple Silicon芯片信息动态配置
+    
+    Returns:
+        Dict[str, Any]: LLM模型参数配置
+    """
+    # 获取Apple Silicon优化配置
+    optimal_config = get_optimal_llama_cpp_config()
+
+    # 基础模型参数
+    model_params = {
+        'streaming': True,
+        'temperature': 0.7,
+        'top_p': 0.9,
+        'top_k': 20,
+        'model_kwargs': {
+            'mini_p': 0,
+            'presence_penalty': 1.5
+        },
+        'verbose': False,
+    }
+
+    # 应用Apple Silicon优化配置
+    model_params.update(optimal_config)
+
+    return model_params
+
+
+def get_apple_silicon_summary() -> Dict[str, Any]:
+    """
+    获取Apple Silicon芯片信息摘要
+    
+    Returns:
+        Dict[str, Any]: 芯片信息摘要
+    """
+    chip_info = get_apple_silicon_info()
+    optimal_config = get_optimal_llama_cpp_config()
+
+    return {
+        'chip_name': chip_info.chip_name,
+        'is_apple_silicon': chip_info.is_apple_silicon,
+        'total_cores': chip_info.total_cores,
+        'performance_cores': chip_info.performance_cores,
+        'efficiency_cores': chip_info.efficiency_cores,
+        'memory_gb': chip_info.memory_gb,
+        'gpu_cores': chip_info.gpu_cores,
+        'optimal_n_threads': optimal_config['n_threads'],
+        'optimal_n_ctx': optimal_config['n_ctx'],
+        'config_note': '仅使用性能核心(P-cores)以获得最佳性能' if chip_info.is_apple_silicon else '标准配置'
+    }
+
+
+# 预设配置模板
+LLAMA_CPP_CONFIG_PRESETS = {
+    'high_performance': {
+        'description': '高性能配置 - 适合Apple M1 Pro/Max及以上',
+        'n_ctx': 8192,
+        'temperature': 0.7,
+        'top_p': 0.9,
+        'top_k': 20,
+    },
+    'balanced': {
+        'description': '平衡配置 - 适合Apple M1/M2基础版',
+        'n_ctx': 4096,
+        'temperature': 0.7,
+        'top_p': 0.9,
+        'top_k': 20,
+    },
+    'memory_efficient': {
+        'description': '内存优化配置 - 适合内存较小的设备',
+        'n_ctx': 2048,
+        'temperature': 0.7,
+        'top_p': 0.9,
+        'top_k': 20,
+    }
+}
+
+
+def get_config_preset(preset_name: str) -> Dict[str, Any]:
+    """
+    获取预设配置
+    
+    Args:
+        preset_name: 预设名称 ('high_performance', 'balanced', 'memory_efficient')
+    
+    Returns:
+        Dict[str, Any]: 预设配置
+    """
+    if preset_name not in LLAMA_CPP_CONFIG_PRESETS:
+        raise ValueError(f"未知的预设配置: {preset_name}")
+
+    preset = LLAMA_CPP_CONFIG_PRESETS[preset_name].copy()
+
+    # 移除描述字段
+    preset.pop('description', None)
+
+    # 添加n_threads配置（基于当前硬件）
+    optimal_config = get_optimal_llama_cpp_config()
+    preset['n_threads'] = optimal_config['n_threads']
+
+    return preset

src/voice_dialogue/services/text/generator.py

@@ -6,6 +6,7 @@ from langchain.memory import ConversationBufferWindowMemory
 from langchain_core.chat_history import InMemoryChatMessageHistory
 
 from voice_dialogue.config import paths
+from voice_dialogue.config.llm_config import get_llm_model_params, get_apple_silicon_summary
 from voice_dialogue.core.base import BaseThread
 from voice_dialogue.core.constants import chat_history_cache
 from voice_dialogue.models.voice_task import VoiceTask, QuestionDisplayMessage
@@ -191,18 +192,17 @@ class LLMResponseGenerator(BaseThread):
 
     def run(self):
         model_path = paths.LLM_MODELS_PATH / 'qwen' / 'Qwen3-8B-Q6_K.gguf'
-        model_params = {
-            'streaming': True,
-            'n_ctx': 32768,
-            'temperature': 0.7,
-            'top_p': 0.9,
-            'top_k': 20,
-            'model_kwargs': {
-                'mini_p': 0,
-                'presence_penalty': 1.5
-            },
-            'verbose': False,
-        }
+        
+        model_params = get_llm_model_params()
+        
+        # 打印芯片信息和优化配置
+        chip_summary = get_apple_silicon_summary()
+        print(f"检测到芯片: {chip_summary['chip_name']}")
+        print(f"性能核心数: {chip_summary['performance_cores']}")
+        print(f"使用线程数: {chip_summary['optimal_n_threads']} (仅使用性能核心)")
+        print(f"上下文窗口: {chip_summary['optimal_n_ctx']}")
+        print(f"配置说明: {chip_summary['config_note']}")
+        
         self.model_instance = create_langchain_chat_llamacpp_instance(
             local_model_path=model_path, model_params=model_params
         )

src/voice_dialogue/utils/apple_silicon.py

@@ -0,0 +1,237 @@
+"""Apple Silicon 芯片信息获取工具"""
+
+import json
+import subprocess
+import platform
+from typing import Dict, Optional
+from dataclasses import dataclass
+
+__all__ = ('AppleSiliconInfo', 'get_apple_silicon_info')
+
+
+@dataclass
+class AppleSiliconInfo:
+    """Apple Silicon 芯片信息"""
+    chip_name: str  # 如 "Apple M1", "Apple M1 Pro", "Apple M2 Max"
+    total_cores: int  # 总核心数
+    performance_cores: int  # 性能核心数（P-cores）
+    efficiency_cores: int  # 效率核心数（E-cores）
+    gpu_cores: int  # GPU核心数
+    memory_gb: int  # 内存大小（GB）
+    is_apple_silicon: bool  # 是否为Apple Silicon
+
+
+def _get_system_profiler_info() -> Optional[Dict]:
+    """通过system_profiler获取硬件信息"""
+    try:
+        # 获取硬件信息
+        result = subprocess.run([
+            'system_profiler', 'SPHardwareDataType', '-json'
+        ], capture_output=True, text=True, timeout=10)
+        
+        if result.returncode == 0:
+            data = json.loads(result.stdout)
+            return data.get('SPHardwareDataType', [{}])[0]
+    except (subprocess.TimeoutExpired, json.JSONDecodeError, FileNotFoundError):
+        pass
+    return None
+
+
+def _parse_cpu_info_from_sysctl() -> Optional[Dict]:
+    """通过sysctl获取CPU信息"""
+    try:
+        # 获取CPU核心信息
+        result = subprocess.run([
+            'sysctl', '-n', 
+            'hw.perflevel0.physicalcpu',  # 性能核心数
+            'hw.perflevel1.physicalcpu',  # 效率核心数
+            'hw.logicalcpu',              # 逻辑核心数
+            'hw.memsize'                  # 内存大小
+        ], capture_output=True, text=True, timeout=5)
+        
+        if result.returncode == 0:
+            lines = result.stdout.strip().split('\n')
+            if len(lines) >= 4:
+                return {
+                    'performance_cores': int(lines[0]) if lines[0].isdigit() else 0,
+                    'efficiency_cores': int(lines[1]) if lines[1].isdigit() else 0,
+                    'total_cores': int(lines[2]) if lines[2].isdigit() else 0,
+                    'memory_bytes': int(lines[3]) if lines[3].isdigit() else 0
+                }
+    except (subprocess.TimeoutExpired, ValueError, IndexError, FileNotFoundError):
+        pass
+    return None
+
+
+def _get_gpu_cores_from_system_profiler() -> int:
+    """获取GPU核心数"""
+    try:
+        result = subprocess.run([
+            'system_profiler', 'SPDisplaysDataType', '-json'
+        ], capture_output=True, text=True, timeout=10)
+        
+        if result.returncode == 0:
+            data = json.loads(result.stdout)
+            displays = data.get('SPDisplaysDataType', [])
+            for display in displays:
+                if 'sppci_cores' in display:
+                    return int(display['sppci_cores'])
+                # 根据芯片名称估算GPU核心数
+                model = display.get('sppci_model', '').lower()
+                if 'm1 max' in model:
+                    return 32
+                elif 'm1 pro' in model:
+                    return 16 if '16-core' in model else 14
+                elif 'm1' in model:
+                    return 8 if '8-core' in model else 7
+                elif 'm2 max' in model:
+                    return 38
+                elif 'm2 pro' in model:
+                    return 19 if '19-core' in model else 16
+                elif 'm2' in model:
+                    return 10 if '10-core' in model else 8
+                elif 'm3 max' in model:
+                    return 40
+                elif 'm3 pro' in model:
+                    return 18 if '18-core' in model else 14
+                elif 'm3' in model:
+                    return 10 if '10-core' in model else 8
+    except (subprocess.TimeoutExpired, json.JSONDecodeError, ValueError, FileNotFoundError):
+        pass
+    return 8  # 默认值
+
+
+def get_apple_silicon_info() -> AppleSiliconInfo:
+    """
+    获取Apple Silicon芯片信息
+    
+    Returns:
+        AppleSiliconInfo: 芯片信息对象
+    """
+    # 检查是否为Apple Silicon
+    machine = platform.machine()
+    is_apple_silicon = machine in ('arm64', 'arm64e') and platform.system() == 'Darwin'
+    
+    if not is_apple_silicon:
+        return AppleSiliconInfo(
+            chip_name="非Apple Silicon",
+            total_cores=1,
+            performance_cores=1,
+            efficiency_cores=0,
+            gpu_cores=0,
+            memory_gb=8,
+            is_apple_silicon=False
+        )
+    
+    # 获取系统信息
+    hw_info = _get_system_profiler_info()
+    cpu_info = _parse_cpu_info_from_sysctl()
+    
+    # 解析芯片名称
+    chip_name = "Apple Silicon"
+    if hw_info and 'chip_type' in hw_info:
+        chip_name = hw_info['chip_type']
+    elif hw_info and 'cpu_type' in hw_info:
+        chip_name = hw_info['cpu_type']
+    
+    # 解析核心信息
+    performance_cores = 4  # 默认值
+    efficiency_cores = 4   # 默认值
+    total_cores = 8        # 默认值
+    
+    if cpu_info:
+        performance_cores = cpu_info.get('performance_cores', performance_cores)
+        efficiency_cores = cpu_info.get('efficiency_cores', efficiency_cores)
+        total_cores = cpu_info.get('total_cores', total_cores)
+    
+    # 如果sysctl失败，尝试从芯片名称推断
+    if performance_cores == 0 and efficiency_cores == 0:
+        chip_lower = chip_name.lower()
+        if 'm1 max' in chip_lower:
+            performance_cores, efficiency_cores = 8, 2
+        elif 'm1 pro' in chip_lower:
+            performance_cores, efficiency_cores = 8, 2
+        elif 'm1' in chip_lower:
+            performance_cores, efficiency_cores = 4, 4
+        elif 'm2 max' in chip_lower:
+            performance_cores, efficiency_cores = 8, 4
+        elif 'm2 pro' in chip_lower:
+            performance_cores, efficiency_cores = 8, 4
+        elif 'm2' in chip_lower:
+            performance_cores, efficiency_cores = 4, 4
+        elif 'm3' in chip_lower:
+            if 'max' in chip_lower:
+                performance_cores, efficiency_cores = 12, 4
+            elif 'pro' in chip_lower:
+                performance_cores, efficiency_cores = 8, 4
+            else:
+                performance_cores, efficiency_cores = 4, 4
+        
+        total_cores = performance_cores + efficiency_cores
+    
+    # 获取内存大小
+    memory_gb = 8  # 默认值
+    if cpu_info and 'memory_bytes' in cpu_info:
+        memory_gb = cpu_info['memory_bytes'] // (1024 ** 3)
+    elif hw_info and 'physical_memory' in hw_info:
+        memory_str = hw_info['physical_memory'].replace(' GB', '').replace(',', '')
+        try:
+            memory_gb = int(memory_str)
+        except ValueError:
+            pass
+    
+    # 获取GPU核心数
+    gpu_cores = _get_gpu_cores_from_system_profiler()
+    
+    return AppleSiliconInfo(
+        chip_name=chip_name,
+        total_cores=total_cores,
+        performance_cores=performance_cores,
+        efficiency_cores=efficiency_cores,
+        gpu_cores=gpu_cores,
+        memory_gb=memory_gb,
+        is_apple_silicon=True
+    )
+
+
+def get_optimal_llama_cpp_config() -> Dict[str, int]:
+    """
+    根据Apple Silicon芯片信息获取最优的llama.cpp配置
+    
+    Returns:
+        Dict[str, int]: 包含n_threads和n_ctx的配置
+    """
+    chip_info = get_apple_silicon_info()
+    
+    if not chip_info.is_apple_silicon:
+        # 非Apple Silicon系统的默认配置
+        return {
+            'n_threads': 4,
+            'n_ctx': 2048
+        }
+    
+    # 对于Apple Silicon，只使用性能核心（P-cores）
+    # 避免混合使用效率核心，以获得最佳性能
+    n_threads = chip_info.performance_cores
+    
+    # 根据内存大小调整上下文窗口
+    if chip_info.memory_gb >= 32:
+        n_ctx = 8192
+    elif chip_info.memory_gb >= 16:
+        n_ctx = 4096
+    else:
+        n_ctx = 2048
+    
+    return {
+        'n_threads': n_threads,
+        'n_ctx': n_ctx
+    }
+
+
+if __name__ == '__main__':
+    # 测试功能
+    info = get_apple_silicon_info()
+    print(f"芯片信息: {info}")
+    
+    config = get_optimal_llama_cpp_config()
+    print(f"推荐配置: {config}") 

src/voice_dialogue/utils/system.py

@@ -5,7 +5,7 @@ import os
 import platform
 from typing import Literal
 
-__all__ = ('get_system_language', 'get_system_info')
+__all__ = ('get_system_language', 'get_system_info', 'is_apple_silicon')
 
 
 def get_system_language() -> Literal['zh', 'en']:
@@ -43,13 +43,24 @@ def get_system_language() -> Literal['zh', 'en']:
                 return 'en'
 
     except Exception:
-        # 如果所有方法都失败，返回默认值
         pass
+        # 如果所有方法都失败，返回默认值
 
     # 默认返回中文
     return 'zh'
 
 
+def is_apple_silicon() -> bool:
+    """
+    检查当前系统是否为Apple Silicon
+    
+    Returns:
+        bool: 如果是Apple Silicon返回True，否则返回False
+    """
+    return (platform.system() == 'Darwin' and
+            platform.machine() in ('arm64', 'arm64e'))
+
+
 def get_system_info() -> dict:
     """
     获取系统信息
@@ -57,7 +68,7 @@ def get_system_info() -> dict:
     Returns:
         dict: 包含系统信息的字典
     """
-    return {
+    info = {
         'platform': platform.system(),
         'platform_version': platform.version(),
         'architecture': platform.architecture()[0],
@@ -65,4 +76,7 @@ def get_system_info() -> dict:
         'processor': platform.processor(),
         'language': get_system_language(),
         'python_version': platform.python_version(),
+        'is_apple_silicon': is_apple_silicon(),
     }
+
+    return info