mirror of
https://github.com/crmeb/CRMEB.git
synced 2026-10-28 21:40:50 +00:00
289 lines
12 KiB
PHP
289 lines
12 KiB
PHP
<?php
|
||
declare(strict_types=1);
|
||
|
||
namespace crmeb\services\ai;
|
||
|
||
/**
|
||
* AI 协议适配器工厂（系统内置）
|
||
*
|
||
* AI 对话能力的唯一入口。协议实现全部内置于 crmeb/services/ai/protocol/，
|
||
* 不再通过 driver/ 目录按厂商可插拔扩展：
|
||
* - 同一协议族内新增厂商 → 后台「AI Agent → 模型管理」加一条模型记录即可，零代码；
|
||
* - 出现新协议族 → 在 protocol/ 下新增 BaseProtocol 子类并在本类 PROTOCOLS 登记。
|
||
*
|
||
* 使用方式：
|
||
* AiProtocol::make('openai_compatible', ['base_url' => ..., 'api_key' => ..., 'model' => ...])->chat('问题');
|
||
*/
|
||
class AiProtocol
|
||
{
|
||
/**
|
||
* 内置协议族清单
|
||
*
|
||
* credential 说明模型记录需要的凭证形态：
|
||
* - model 每条模型自带 api_key
|
||
* - global 复用全局系统配置凭证，模型记录无需填写
|
||
*
|
||
* @var array
|
||
*/
|
||
protected const PROTOCOLS = [
|
||
'openai_compatible' => [
|
||
'class' => \crmeb\services\ai\protocol\OpenAiCompatible::class,
|
||
'name' => 'OpenAI 兼容协议',
|
||
'credential' => 'model',
|
||
'base_url_required' => true,
|
||
'default_base_url' => '',
|
||
'description' => '适用于 DeepSeek、通义千问、Kimi、智谱 GLM、Moonshot、豆包、OpenAI 及 vLLM/Ollama 等自建网关',
|
||
],
|
||
'yihaotong' => [
|
||
'class' => \crmeb\services\ai\protocol\Yihaotong::class,
|
||
'name' => '一号通 AI',
|
||
'credential' => 'global',
|
||
'base_url_required' => false,
|
||
'default_base_url' => '',
|
||
'description' => '复用后台「一号通设置」的全局凭证，模型记录只需填写模型标识',
|
||
],
|
||
];
|
||
|
||
/**
|
||
* 获取全部内置协议定义（供后台模型管理表单渲染）
|
||
* @return array 以协议标识为键的定义数组
|
||
*/
|
||
public static function definitions(): array
|
||
{
|
||
return self::PROTOCOLS;
|
||
}
|
||
|
||
/**
|
||
* 获取协议下拉选项（供前端选择器使用）
|
||
* @return array [['protocol' => string, 'name' => string, 'credential' => string, 'base_url_required' => bool, 'description' => string]]
|
||
*/
|
||
public static function options(): array
|
||
{
|
||
$options = [];
|
||
foreach (self::PROTOCOLS as $protocol => $meta) {
|
||
$options[] = [
|
||
'protocol' => $protocol,
|
||
'name' => $meta['name'],
|
||
'credential' => $meta['credential'],
|
||
'base_url_required' => $meta['base_url_required'],
|
||
'description' => $meta['description'],
|
||
];
|
||
}
|
||
return $options;
|
||
}
|
||
|
||
/**
|
||
* 判断协议标识是否受支持
|
||
* @param string $protocol 协议标识
|
||
* @return bool
|
||
*/
|
||
public static function has(string $protocol): bool
|
||
{
|
||
return isset(self::PROTOCOLS[$protocol]);
|
||
}
|
||
|
||
/**
|
||
* 读取协议展示名称
|
||
* @param string $protocol 协议标识
|
||
* @return string 未登记的协议返回标识本身
|
||
*/
|
||
public static function name(string $protocol): string
|
||
{
|
||
return (string) (self::PROTOCOLS[$protocol]['name'] ?? $protocol);
|
||
}
|
||
|
||
/**
|
||
* 创建协议适配器实例
|
||
* @param string $protocol 协议标识（openai_compatible / yihaotong）
|
||
* @param array $config 模型配置键值对
|
||
* @return AiInterface 已注入配置的适配器实例
|
||
* @throws \InvalidArgumentException 协议不受支持时抛出
|
||
*/
|
||
public static function make(string $protocol, array $config = []): AiInterface
|
||
{
|
||
if (!self::has($protocol)) {
|
||
throw new \InvalidArgumentException('不支持的 AI 协议：' . $protocol);
|
||
}
|
||
|
||
$class = (string) self::PROTOCOLS[$protocol]['class'];
|
||
/** @var AiInterface $handler */
|
||
$handler = new $class();
|
||
$handler->setConfig($config);
|
||
return $handler;
|
||
}
|
||
|
||
/**
|
||
* 流式对话（按适配器能力自动降级）
|
||
*
|
||
* 适配器实现 StreamAiInterface 时走真流式逐块回调，
|
||
* 否则发起非流式请求后一次性回调，调用方无需关心协议差异。
|
||
*
|
||
* @param AiInterface $handler 适配器实例
|
||
* @param string $message 对话内容
|
||
* @param callable $onChunk 增量回调 function(array $delta): void，$delta = ['type' => 'reasoning'|'content', 'text' => string]
|
||
* @param array $options 附加选项（prompts: 提示词一维数组）
|
||
*/
|
||
public static function chatStream(AiInterface $handler, string $message, callable $onChunk, array $options = []): void
|
||
{
|
||
// 内嵌思维链过滤器：部分推理模型不返回标准 reasoning 字段，而是把思维链以
|
||
// <think>...</think>（含 </think_never_used_xxx> 等变体闭合标签）内嵌在正文流中，
|
||
// 在唯一出口统一改道/剥离，避免内部过程与标签混入最终答复（如编辑器 AI 写作场景）。
|
||
$state = self::newThinkFilterState();
|
||
|
||
if ($handler instanceof StreamAiInterface) {
|
||
$handler->chatStream($message, function (array $delta) use ($onChunk, &$state): void {
|
||
if (($delta['type'] ?? 'content') !== 'reasoning') {
|
||
self::filterThinkChunk($state, (string) ($delta['text'] ?? ''), $onChunk);
|
||
return;
|
||
}
|
||
$onChunk($delta);
|
||
}, $options);
|
||
self::flushThinkFilter($state, $onChunk);
|
||
return;
|
||
}
|
||
|
||
$result = $handler->chat($message, $options + ['stream' => 0]);
|
||
if (!empty($result['reasoning'])) {
|
||
$onChunk(['type' => 'reasoning', 'text' => (string) $result['reasoning']]);
|
||
}
|
||
if (($result['content'] ?? '') !== '') {
|
||
self::filterThinkChunk($state, (string) $result['content'], $onChunk);
|
||
self::flushThinkFilter($state, $onChunk);
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 创建内嵌思维链流式过滤器状态
|
||
* @return array 过滤器状态（引用传递给 filterThinkChunk / flushThinkFilter）
|
||
*/
|
||
private static function newThinkFilterState(): array
|
||
{
|
||
return [
|
||
'inThink' => false, // 当前是否位于 <think> 标签内部
|
||
'pending' => '', // 跨块截断的标签前缀缓冲
|
||
'hasContent' => false, // 正文是否已开始外发（用于判定孤立闭合标记前的文本归属）
|
||
];
|
||
}
|
||
|
||
/**
|
||
* 过滤器逐块处理正文增量
|
||
*
|
||
* 1. <think>...</think> 完整思维链：标签内文本改道为 reasoning 增量；
|
||
* 2. 漏发开标签、仅输出 </think_xxx> 变体闭合标记：正文开始前的文本实为思维链，改道 reasoning；
|
||
* 正文已开始后出现的孤立闭合标记视为噪声，仅剥离标记本身；
|
||
* 3. 标签可能跨块截断，未判定完整性的尾部短前缀暂存缓冲。
|
||
*
|
||
* @param array $state 过滤器状态
|
||
* @param string $text 本次到达的正文增量
|
||
* @param callable $onChunk 增量回调 function(array $delta): void，$delta = ['type' => 'reasoning'|'content', 'text' => string]
|
||
*/
|
||
private static function filterThinkChunk(array &$state, string $text, callable $onChunk): void
|
||
{
|
||
if ($text === '') {
|
||
return;
|
||
}
|
||
$buffer = $state['pending'] . $text;
|
||
$state['pending'] = '';
|
||
|
||
while ($buffer !== '') {
|
||
if ($state['inThink']) {
|
||
// 思维链模式：扫描闭合标签（兼容 </think_xxx> 变体）
|
||
$matched = preg_match('~</think[^>]*>~i', $buffer, $m, PREG_OFFSET_CAPTURE) ? $m[0] : null;
|
||
if ($matched === null) {
|
||
$hold = self::thinkHoldSuffix($buffer);
|
||
$emit = substr($buffer, 0, strlen($buffer) - strlen($hold));
|
||
if ($emit !== '') {
|
||
$onChunk(['type' => 'reasoning', 'text' => $emit]);
|
||
}
|
||
$state['pending'] = $hold;
|
||
return;
|
||
}
|
||
$pos = (int) $matched[1];
|
||
if ($pos > 0) {
|
||
$onChunk(['type' => 'reasoning', 'text' => substr($buffer, 0, $pos)]);
|
||
}
|
||
$buffer = substr($buffer, $pos + strlen((string) $matched[0]));
|
||
$state['inThink'] = false;
|
||
continue;
|
||
}
|
||
|
||
// 正文模式：扫描 <think> 开标签
|
||
$matched = preg_match('~<think[^>]*>~i', $buffer, $m, PREG_OFFSET_CAPTURE) ? $m[0] : null;
|
||
// 部分兼容网关会漏掉开标签、只输出 </think_xxx>。闭合标记本身不能进入最终答复。
|
||
$closing = preg_match('~</think[^>]*>~i', $buffer, $close, PREG_OFFSET_CAPTURE) ? $close[0] : null;
|
||
if ($closing !== null && ($matched === null || (int) $closing[1] < (int) $matched[1])) {
|
||
$pos = (int) $closing[1];
|
||
$pre = substr($buffer, 0, $pos);
|
||
if ($state['hasContent']) {
|
||
// 正文已开始：前置文本仍是正文，仅剥离标记
|
||
if ($pre !== '') {
|
||
$onChunk(['type' => 'content', 'text' => $pre]);
|
||
}
|
||
} elseif (trim($pre) !== '') {
|
||
// 正文尚未开始：标记前的文本实为思维链，改道 reasoning
|
||
$onChunk(['type' => 'reasoning', 'text' => $pre]);
|
||
$state['hasContent'] = true;
|
||
}
|
||
$buffer = substr($buffer, $pos + strlen((string) $closing[0]));
|
||
continue;
|
||
}
|
||
if ($matched === null) {
|
||
$hold = self::thinkHoldSuffix($buffer);
|
||
$emit = substr($buffer, 0, strlen($buffer) - strlen($hold));
|
||
if ($emit !== '') {
|
||
$onChunk(['type' => 'content', 'text' => $emit]);
|
||
$state['hasContent'] = true;
|
||
}
|
||
$state['pending'] = $hold;
|
||
return;
|
||
}
|
||
$pos = (int) $matched[1];
|
||
if ($pos > 0) {
|
||
$onChunk(['type' => 'content', 'text' => substr($buffer, 0, $pos)]);
|
||
$state['hasContent'] = true;
|
||
}
|
||
$buffer = substr($buffer, $pos + strlen((string) $matched[0]));
|
||
$state['inThink'] = true;
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 流结束后冲刷过滤器残留缓冲（跨块截断的未闭合标签前缀）
|
||
* @param array $state 过滤器状态
|
||
* @param callable $onChunk 增量回调 function(array $delta): void
|
||
*/
|
||
private static function flushThinkFilter(array &$state, callable $onChunk): void
|
||
{
|
||
$pending = (string) $state['pending'];
|
||
if ($pending === '') {
|
||
return;
|
||
}
|
||
$state['pending'] = '';
|
||
if ($state['inThink']) {
|
||
$onChunk(['type' => 'reasoning', 'text' => $pending]);
|
||
return;
|
||
}
|
||
$onChunk(['type' => 'content', 'text' => $pending]);
|
||
$state['hasContent'] = true;
|
||
}
|
||
|
||
/**
|
||
* 提取缓冲尾部可能是跨块截断标签前缀的短后缀（自最后一个未闭合的 "<" 起）
|
||
* @param string $buffer 当前缓冲
|
||
* @return string 需暂存待下一块合并判定的后缀，无需暂存时返回空串
|
||
*/
|
||
private static function thinkHoldSuffix(string $buffer): string
|
||
{
|
||
$lastLt = strrpos($buffer, '<');
|
||
if ($lastLt === false) {
|
||
return '';
|
||
}
|
||
$suffix = substr($buffer, $lastLt);
|
||
// 已闭合（含 >）说明不是跨块截断的标签前缀
|
||
if (strpos($suffix, '>') !== false) {
|
||
return '';
|
||
}
|
||
return $suffix;
|
||
}
|
||
}
|