shuiyu67andWorkBuddy
5fe1a7c868
- openai/ollama/manual 三引擎均暴露 max_input_tokens 属性（统一契约） - ollama 额外上报 prompt_eval_count 到 last_prompt_tokens，使本地模型也走 token 感知压缩 - should_compress 改用 max_input_tokens，系统依当前引擎上限的 70%（compress_ratio）自动触发 - 压缩始终调用当前引擎（current_engine）做总结，符合"调用当前引擎总结压缩" - 新增 tests/test_max_input_tokens.py 覆盖契约 + 70% 触发 + 消息数兜底 Co-Authored-By: WorkBuddy <workbuddy@tencent.com>
537 lines
21 KiB
Python
537 lines
21 KiB
Python
import os
|
||
import re
|
||
import sys
|
||
import json
|
||
import requests
|
||
import threading
|
||
import time
|
||
from colorama import Fore, Style
|
||
|
||
# 延迟导入 ollama，避免程序启动时卡住
|
||
# 如果导入失败，ollama_client 将为 None
|
||
try:
|
||
import ollama
|
||
ollama_client = ollama
|
||
except Exception as e:
|
||
print(f"{Fore.YELLOW}警告: 无法导入 ollama 库: {e}{Style.RESET_ALL}")
|
||
print(f"{Fore.YELLOW}请确保已安装 ollama 包 (pip install ollama){Style.RESET_ALL}")
|
||
ollama_client = None
|
||
|
||
# 添加项目根目录到sys.path
|
||
current_dir = os.path.dirname(os.path.abspath(__file__))
|
||
project_root = os.path.dirname(current_dir)
|
||
sys.path.insert(0, project_root)
|
||
|
||
try:
|
||
from xcli_core.verbose import vprint as _vprint
|
||
except Exception:
|
||
def _vprint(*args, **kwargs):
|
||
pass
|
||
|
||
# <input> 特殊语法解析（配置模型时可交互式询问用户）
|
||
try:
|
||
from xcli_core.config import resolve_input_value
|
||
except Exception:
|
||
def resolve_input_value(value, field_name="", default_prompt=None):
|
||
return value
|
||
|
||
try:
|
||
from xcli_core.tool_result import normalize_tool_text
|
||
except Exception:
|
||
def normalize_tool_text(result, default="无结果"):
|
||
"""兜底归一化：ToolResult / dict / str 统一成字符串"""
|
||
if result is None:
|
||
return default
|
||
if isinstance(result, dict):
|
||
return str(result.get("result", default))
|
||
return str(result)
|
||
|
||
|
||
class OllamaAI:
|
||
"""基于Ollama的本地AI引擎"""
|
||
requires_api_key = False # Ollama是本地引擎，不需要API密钥
|
||
|
||
def get_help_info(self):
|
||
"""获取引擎帮助信息"""
|
||
help_text = """
|
||
Ollama AI引擎插件帮助信息
|
||
========================
|
||
|
||
引擎名称: Ollama
|
||
描述: 基于Ollama的本地AI引擎,支持多种大型语言模型
|
||
|
||
可用命令:
|
||
/engine.ollama models - 列出本地可用的模型
|
||
/engine.ollama remote-models - 显示可下载的模型列表
|
||
/engine.ollama search <关键词> - 搜索可下载的模型
|
||
/engine.ollama set <模型名> - 设置当前使用的模型
|
||
/engine.ollama pull <模型名> - 从Ollama库拉取模型
|
||
|
||
注意: 本引擎强制使用流式输出，AI回复会实时显示。
|
||
|
||
使用说明:
|
||
1. 确保Ollama服务已在本地运行
|
||
2. 使用/engine.switch ollama切换到此引擎
|
||
3. 使用上述命令管理模型和与AI交互
|
||
|
||
思考模式:
|
||
- 支持深度思考模型（如 QwQ、DeepSeek-R1）
|
||
- 思考过程会自动以灰色文本显示
|
||
"""
|
||
return help_text
|
||
|
||
def __init__(self):
|
||
self.name = "ollama"
|
||
self.cli = None # 引用CLI实例以访问插件
|
||
# 初始化对话历史
|
||
self.conversation_history = []
|
||
|
||
# 直接从 config.json 读取配置
|
||
# 模型最大 token 大小（单位 K，如 32 = 32K = 32768 token）
|
||
# 支持 <input> 语法：配置里写 <input:提示语> 会在配置模型时询问用户
|
||
max_token_k = "<input:请输入模型最大token大小(K)>"
|
||
config_file = os.path.join(project_root, "config.json")
|
||
try:
|
||
with open(config_file, 'r', encoding='utf-8') as f:
|
||
config = json.load(f)
|
||
engine_config = config.get("api", {}).get("engines", {}).get("ollama", {})
|
||
self.api_key = engine_config.get("api_key", "")
|
||
self.base_url = engine_config.get("base_url", "")
|
||
self.model = engine_config.get("model", "")
|
||
self.max_history = config.get("system", {}).get("max_history", 10)
|
||
max_token_k = engine_config.get("max_token_k", max_token_k)
|
||
except Exception as e:
|
||
# 如果读取失败，使用空值
|
||
self.api_key = ""
|
||
self.base_url = ""
|
||
self.model = ""
|
||
self.max_history = 10
|
||
|
||
# 解析 <input> 语法（询问用户）并换算成 num_ctx（token 数）
|
||
resolved_k = resolve_input_value(max_token_k, field_name="模型最大token大小(K)")
|
||
self.num_ctx = self._parse_num_ctx(resolved_k)
|
||
|
||
# 共享对话历史引用（由CLI设置）
|
||
self.shared_conversation_history = None
|
||
|
||
# 最近一次请求的 prompt token 数（供 token 感知压缩使用）
|
||
self.last_prompt_tokens = None
|
||
|
||
# 直接启用思考标记，由主程序自动处理深度思考模型的响应
|
||
self.thinking_start_marker = "<thinking>"
|
||
self.thinking_end_marker = "</thinking>"
|
||
|
||
# 检测Ollama服务是否运行
|
||
self.is_service_running = self.check_ollama_service()
|
||
|
||
if self.is_service_running:
|
||
_vprint(f"{Fore.GREEN}Ollama服务检测成功,引擎已启用{Style.RESET_ALL}")
|
||
else:
|
||
print(f"{Fore.RED}Ollama服务未运行,引擎将不可用{Style.RESET_ALL}")
|
||
|
||
@staticmethod
|
||
def _parse_num_ctx(value):
|
||
"""把用户输入的 K 值换算成 token 数：32 / 32K → 32768。非法/空 → 4096"""
|
||
if value is None:
|
||
return 4096
|
||
s = str(value).strip().lower()
|
||
digits = re.sub(r"[^\d.]", "", s)
|
||
if not digits:
|
||
return 4096
|
||
try:
|
||
k = float(digits)
|
||
except ValueError:
|
||
return 4096
|
||
if k <= 0:
|
||
return 4096
|
||
return max(1024, int(k * 1024))
|
||
|
||
def check_ollama_service(self):
|
||
"""检测Ollama服务是否运行"""
|
||
# 检查 ollama 库是否可用
|
||
if ollama_client is None:
|
||
return False
|
||
try:
|
||
# 尝试调用Ollama API获取版本信息
|
||
response = ollama_client.list()
|
||
# 如果没有异常,说明服务运行正常
|
||
return True
|
||
except Exception as e:
|
||
# 如果出现异常,说明服务未运行
|
||
return False
|
||
|
||
@property
|
||
def max_input_tokens(self):
|
||
"""当前引擎支持的最大输入 token 数（统一 API，供上下文压缩使用）。
|
||
本地 Ollama 由配置 max_token_k 换算的 num_ctx 决定。"""
|
||
return self.num_ctx
|
||
|
||
def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
|
||
"""
|
||
生成AI响应 - 使用Ollama API
|
||
|
||
Args:
|
||
user_input: 用户输入
|
||
tool_results: 工具执行结果
|
||
system_prompt: 系统提示词
|
||
tools: OpenAI FC 工具定义列表 (可选)
|
||
"""
|
||
try:
|
||
# 检查Ollama服务是否运行
|
||
if not self.is_service_running:
|
||
return "错误:Ollama服务未运行,请启动Ollama服务后再使用此引擎."
|
||
|
||
# 检查模型是否已配置
|
||
if not self.model or not self.model.strip():
|
||
return ("错误:Ollama模型未配置。请执行以下操作:\n"
|
||
" 1. /engine.ollama models → 查看本地可用模型\n"
|
||
" 2. /engine.ollama set <模型名> → 设置要使用的模型\n"
|
||
" 或编辑 config.json 中 api.engines.ollama.model 字段")
|
||
|
||
# 如果有工具结果,将其作为上下文返回给AI处理
|
||
if tool_results:
|
||
tool_result_text = f"工具执行结果: {normalize_tool_text(tool_results)}"
|
||
messages = self._build_messages_with_history(system_prompt, tool_result_text)
|
||
else:
|
||
messages = self._build_messages_with_history(system_prompt, user_input)
|
||
|
||
# 构建调用参数（num_ctx: 模型上下文窗口大小，由配置的 max_token_k 决定）
|
||
kwargs = {
|
||
"model": self.model,
|
||
"messages": messages,
|
||
"stream": False,
|
||
"options": {"num_ctx": self.num_ctx}
|
||
}
|
||
|
||
# ── Function Calling: 注入工具定义 ──
|
||
if tools:
|
||
kwargs["tools"] = tools
|
||
|
||
# 使用Ollama Python库调用API
|
||
response = ollama_client.chat(**kwargs)
|
||
|
||
# ── Function Calling: 检查 tool_calls ──
|
||
message = response.get("message", {})
|
||
# 上报本轮 prompt token 数（Ollama 返回 prompt_eval_count），供 token 感知压缩
|
||
try:
|
||
if isinstance(response, dict):
|
||
_pet = response.get("prompt_eval_count")
|
||
else:
|
||
_pet = getattr(response, "prompt_eval_count", None)
|
||
self.last_prompt_tokens = int(_pet) if _pet else None
|
||
except Exception:
|
||
self.last_prompt_tokens = None
|
||
tool_calls_raw = message.get("tool_calls")
|
||
if tool_calls_raw:
|
||
# Ollama 的 tool_calls 格式: [{"function": {"name": ..., "arguments": {...}}}]
|
||
fc_calls = []
|
||
for i, tc in enumerate(tool_calls_raw):
|
||
func = tc.get("function", {})
|
||
name = func.get("name", "")
|
||
args = func.get("arguments", {})
|
||
call_id = f"call_{name}_{i}"
|
||
|
||
# 统一为小狸内部格式
|
||
if "args" in args and isinstance(args["args"], str):
|
||
final_args = args["args"]
|
||
elif "args" in args:
|
||
final_args = json.dumps(args["args"], ensure_ascii=False)
|
||
else:
|
||
parts = []
|
||
for k, v in args.items():
|
||
parts.append(str(v) if isinstance(v, str) else json.dumps(v, ensure_ascii=False))
|
||
final_args = " ".join(parts) if parts else ""
|
||
|
||
fc_calls.append({
|
||
"id": call_id,
|
||
"name": name,
|
||
"arguments": args,
|
||
"action": "use_tool",
|
||
"tool": name,
|
||
"args": final_args,
|
||
})
|
||
|
||
if fc_calls:
|
||
return json.dumps({
|
||
"_fc": True,
|
||
"tool_calls": fc_calls,
|
||
}, ensure_ascii=False)
|
||
|
||
# 普通文本响应
|
||
content = message.get("content", "")
|
||
# 清理无效的 UTF-8 代理字符
|
||
content = content.encode('utf-8', 'replace').decode('utf-8')
|
||
return content
|
||
|
||
except Exception as e:
|
||
return f"Ollama API调用失败: {str(e)}"
|
||
|
||
def _build_messages_with_history(self, system_prompt, user_input):
|
||
"""构建包含历史记录的消息列表（支持 FC tool 消息）"""
|
||
messages = []
|
||
|
||
# 添加系统提示词
|
||
if system_prompt:
|
||
clean_prompt = system_prompt.encode('utf-8', 'replace').decode('utf-8')
|
||
messages.append({
|
||
"role": "system",
|
||
"content": clean_prompt
|
||
})
|
||
|
||
# 添加历史对话记录（使用共享历史）
|
||
if self.shared_conversation_history is not None:
|
||
for msg in self.shared_conversation_history[-self.max_history:]:
|
||
if not isinstance(msg, dict) or "role" not in msg:
|
||
continue
|
||
|
||
role = msg["role"]
|
||
|
||
# tool 角色消息（FC 工具结果）—— 官方格式不含 name 字段
|
||
if role == "tool":
|
||
tool_msg = {"role": "tool", "content": msg.get("content", "")}
|
||
if "tool_call_id" in msg:
|
||
tool_msg["tool_call_id"] = msg["tool_call_id"]
|
||
messages.append(tool_msg)
|
||
continue
|
||
|
||
# assistant 消息可能包含 tool_calls
|
||
if role == "assistant" and "tool_calls" in msg:
|
||
assistant_msg = {
|
||
"role": "assistant",
|
||
"content": msg.get("content"),
|
||
"tool_calls": msg["tool_calls"]
|
||
}
|
||
messages.append(assistant_msg)
|
||
continue
|
||
|
||
# 普通 user/assistant 消息
|
||
if "content" in msg:
|
||
clean_content = msg["content"]
|
||
if isinstance(clean_content, str):
|
||
clean_content = clean_content.encode('utf-8', 'replace').decode('utf-8')
|
||
messages.append({"role": role, "content": clean_content})
|
||
|
||
# 清理用户输入
|
||
if user_input:
|
||
clean_input = user_input.encode('utf-8', 'replace').decode('utf-8')
|
||
messages.append({
|
||
"role": "user",
|
||
"content": clean_input
|
||
})
|
||
|
||
return messages
|
||
|
||
|
||
|
||
def _call_api_with_history(self, content, system_prompt=None):
|
||
"""使用对话历史调用API"""
|
||
if not self.is_service_running:
|
||
return "错误:Ollama服务未运行,请启动Ollama服务后再使用此引擎."
|
||
|
||
# 构建消息列表
|
||
messages = []
|
||
# 添加系统提示词(如果提供)
|
||
if system_prompt:
|
||
messages.append({
|
||
"role": "system",
|
||
"content": system_prompt
|
||
})
|
||
# 添加历史对话记录（使用共享历史）
|
||
if self.shared_conversation_history is not None:
|
||
# 确保历史记录格式正确
|
||
for msg in self.shared_conversation_history[-self.max_history:]:
|
||
if isinstance(msg, dict) and "role" in msg and "content" in msg:
|
||
messages.append(msg)
|
||
else:
|
||
# 如果格式不正确，跳过该条记录
|
||
continue
|
||
# 添加当前内容
|
||
messages.append({
|
||
"role": "tool",
|
||
"content": content
|
||
})
|
||
|
||
# 使用Ollama Python库调用API
|
||
response = ollama_client.chat(
|
||
model=self.model,
|
||
messages=messages,
|
||
stream=False
|
||
)
|
||
|
||
return response["message"]["content"]
|
||
|
||
def set_model(self, model_name):
|
||
"""设置要使用的模型"""
|
||
if not self.is_service_running:
|
||
print(f"{Fore.RED}错误:Ollama服务未运行,无法切换模型{Style.RESET_ALL}")
|
||
return
|
||
|
||
self.model = model_name
|
||
print(f"{Fore.GREEN}已切换Ollama模型为: {model_name}{Style.RESET_ALL}")
|
||
|
||
def list_models(self):
|
||
"""列出本地可用的模型"""
|
||
if not self.is_service_running:
|
||
return ["错误:Ollama服务未运行"]
|
||
|
||
try:
|
||
response = ollama_client.list()
|
||
models = [model["name"] for model in response["models"]]
|
||
return models
|
||
except Exception as e:
|
||
return [f"获取模型列表失败: {str(e)}"]
|
||
|
||
def list_remote_models(self, limit=10):
|
||
"""获取可下载的模型列表(仅展示前N个)"""
|
||
if not self.is_service_running:
|
||
return ["错误:Ollama服务未运行"]
|
||
|
||
try:
|
||
# 使用Ollama Python库获取可下载的模型列表
|
||
# 注意:Ollama库本身没有直接获取可下载模型列表的API
|
||
# 这里返回一些常见的模型作为示例
|
||
# 实际应用中可能需要从Ollama官网或其他来源获取此信息
|
||
common_models = [
|
||
"llama3.1:latest",
|
||
"llama3:latest",
|
||
"mistral:latest",
|
||
"gemma2:latest",
|
||
"phi3:latest",
|
||
"qwen2:latest",
|
||
"command-r:latest",
|
||
"dolphin-mistral:latest",
|
||
"nous-hermes2:latest",
|
||
"llava:latest",
|
||
"codellama:latest",
|
||
"mxbai-embed-large:latest",
|
||
"all-minilm:latest",
|
||
"stable-diffusion:latest",
|
||
"moondream:latest"
|
||
]
|
||
return common_models[:limit]
|
||
except Exception as e:
|
||
return [f"获取远程模型列表失败: {str(e)}"]
|
||
|
||
def search_models(self, query, limit=10):
|
||
"""模糊搜索可下载的模型"""
|
||
if not self.is_service_running:
|
||
return ["错误:Ollama服务未运行"]
|
||
|
||
try:
|
||
# 使用Ollama Python库获取可下载的模型列表
|
||
# 注意:Ollama库本身没有直接获取可下载模型列表的API
|
||
all_models = [
|
||
"llama3.1:latest", "llama3:latest", "llama2:latest",
|
||
"mistral:latest", "mistral-openorca:latest",
|
||
"gemma2:latest", "gemma:latest",
|
||
"phi3:latest", "phi2:latest",
|
||
"qwen2:latest", "qwen:latest",
|
||
"command-r:latest", "command-r-plus:latest",
|
||
"dolphin-mistral:latest", "dolphin-llama3:latest",
|
||
"nous-hermes2:latest", "nous-hermes:latest",
|
||
"llava:latest", "bakllava:latest",
|
||
"codellama:latest", "codeup:latest",
|
||
"mxbai-embed-large:latest", "nomic-embed-text:latest",
|
||
"all-minilm:latest", "bge-m3:latest",
|
||
"stable-diffusion:latest", "moondream:latest",
|
||
"llama3-groq:latest", "eva-qwen2:latest"
|
||
]
|
||
|
||
# 进行模糊搜索(简单匹配,包含查询词的模型)
|
||
matched_models = [model for model in all_models if query.lower() in model.lower()]
|
||
|
||
# 限制返回数量
|
||
return matched_models[:limit]
|
||
except Exception as e:
|
||
return [f"搜索模型失败: {str(e)}"]
|
||
|
||
def pull_model(self, model_name):
|
||
"""从Ollama库拉取模型"""
|
||
if not self.is_service_running:
|
||
print(f"{Fore.RED}错误:Ollama服务未运行,无法拉取模型{Style.RESET_ALL}")
|
||
return False
|
||
|
||
try:
|
||
print(f"{Fore.YELLOW}正在拉取模型 {model_name}...{Style.RESET_ALL}")
|
||
|
||
# 使用Ollama Python库拉取模型
|
||
response = ollama_client.pull(model_name, stream=True)
|
||
|
||
# 逐块处理响应
|
||
for part in response:
|
||
if 'status' in part:
|
||
status = part['status']
|
||
print(f"{Fore.YELLOW}{status}{Style.RESET_ALL}")
|
||
# 检查是否完成
|
||
if 'completed' in part and 'total' in part:
|
||
if part['completed'] == part['total']:
|
||
print(f"{Fore.GREEN}模型 {model_name} 拉取完成!{Style.RESET_ALL}")
|
||
|
||
print(f"{Fore.GREEN}模型 {model_name} 拉取成功!{Style.RESET_ALL}")
|
||
return True
|
||
except Exception as e:
|
||
print(f"{Fore.RED}拉取模型 {model_name} 失败: {str(e)}{Style.RESET_ALL}")
|
||
return False
|
||
|
||
def handle_command(self, command):
|
||
"""处理引擎特定命令"""
|
||
if not self.is_service_running:
|
||
print(f"{Fore.RED}Ollama服务未运行,无法执行命令{Style.RESET_ALL}")
|
||
return False
|
||
|
||
if command == "models":
|
||
# 显示本地可用模型
|
||
models = self.list_models()
|
||
print(f"{Fore.GREEN}本地可用的Ollama模型:{Style.RESET_ALL}")
|
||
for model in models:
|
||
status = " (当前使用)" if model == self.model else ""
|
||
print(f"{Fore.GREEN} - {model}{status}{Style.RESET_ALL}")
|
||
return True
|
||
|
||
elif command == "remote-models":
|
||
# 显示可下载的模型(前10个)
|
||
models = self.list_remote_models()
|
||
print(f"{Fore.GREEN}可下载的Ollama模型(前10个):{Style.RESET_ALL}")
|
||
for i, model in enumerate(models, 1):
|
||
print(f"{Fore.GREEN} {i}. {model}{Style.RESET_ALL}")
|
||
return True
|
||
|
||
elif command.startswith("search "):
|
||
# 搜索模型
|
||
query = command[7:].strip() # 移除 'search '
|
||
if query:
|
||
models = self.search_models(query)
|
||
if models:
|
||
print(f"{Fore.GREEN}搜索 '{query}' 的结果(前10个):{Style.RESET_ALL}")
|
||
for i, model in enumerate(models, 1):
|
||
print(f"{Fore.GREEN} {i}. {model}{Style.RESET_ALL}")
|
||
else:
|
||
print(f"{Fore.YELLOW}未找到与 '{query}' 匹配的模型{Style.RESET_ALL}")
|
||
return True
|
||
else:
|
||
print(f"{Fore.RED}请提供搜索关键词.用法: /engine.search <关键词>{Style.RESET_ALL}")
|
||
return False
|
||
|
||
elif command.startswith("set "):
|
||
# 设置模型
|
||
model_name = command[4:].strip() # 移除 'set '
|
||
if model_name:
|
||
self.set_model(model_name)
|
||
return True
|
||
else:
|
||
print(f"{Fore.RED}请提供模型名称.用法: /engine.set <模型名>{Style.RESET_ALL}")
|
||
return False
|
||
|
||
elif command.startswith("pull "):
|
||
# 拉取模型
|
||
model_name = command[5:].strip() # 移除 'pull '
|
||
if model_name:
|
||
success = self.pull_model(model_name)
|
||
return success
|
||
else:
|
||
print(f"{Fore.RED}请提供模型名称.用法: /engine.pull <模型名>{Style.RESET_ALL}")
|
||
return False
|
||
|
||
else:
|
||
print(f"{Fore.RED}未知的Ollama命令.可用命令: models, remote-models, search, set, pull{Style.RESET_ALL}")
|
||
return False |