feat(aichat): 添加/gpt命令,直接聊天 (#1190)

*  添加大模型聊天,便于使用版本的

* 🐛 减少重复关键词

* 🎨 优化换行符

*  添加转换函数

*  添加lint优化

* 🎨 按字符切片

* 🎨 修改lint
This commit is contained in:
himawari 2025-09-01 22:38:15 +08:00 committed by GitHub
parent 1f66f47ce6
commit 20d49ccf15
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -2,6 +2,7 @@
package aichat
import (
"errors"
"math/rand"
"strconv"
"strings"
@ -46,7 +47,9 @@ var (
"- 设置AI聊天(不)以AI语音输出\n" +
"- 查看AI聊天配置\n" +
"- 重置AI聊天\n" +
"- 群聊总结 [消息数目]|群聊总结 1000\n",
"- 群聊总结 [消息数目]|群聊总结 1000\n" +
"- /gpt [内容] (使用大模型聊天)\n",
PrivateDataFolder: "aichat",
})
)
@ -61,6 +64,32 @@ var (
limit = ctxext.NewLimiterManager(time.Second*30, 1)
)
// getModelParams 获取模型参数:温度(float32(temp)/100)、TopP和最大长度
func getModelParams(temp int64) (temperature float32, topp float32, maxn uint) {
// 处理温度参数
if temp <= 0 {
temp = 70 // default setting
}
if temp > 100 {
temp = 100
}
temperature = float32(temp) / 100
// 处理TopP参数
topp = cfg.TopP
if topp == 0 {
topp = 0.9
}
// 处理最大长度参数
maxn = cfg.MaxN
if maxn == 0 {
maxn = 4096
}
return temperature, topp, maxn
}
func init() {
en.OnMessage(ensureconfig, func(ctx *zero.Ctx) bool {
return ctx.ExtractPlainText() != "" &&
@ -88,39 +117,25 @@ func init() {
return
}
if temp <= 0 {
temp = 70 // default setting
}
if temp > 100 {
temp = 100
}
temperature, topp, maxn := getModelParams(temp)
x := deepinfra.NewAPI(cfg.API, cfg.Key)
var mod model.Protocol
maxn := cfg.MaxN
if maxn == 0 {
maxn = 4096
}
topp := cfg.TopP
if topp == 0 {
topp = 0.9
}
switch cfg.Type {
case 0:
mod = model.NewOpenAI(
cfg.ModelName, cfg.Separator,
float32(temp)/100, topp, maxn,
temperature, topp, maxn,
)
case 1:
mod = model.NewOLLaMA(
cfg.ModelName, cfg.Separator,
float32(temp)/100, topp, maxn,
temperature, topp, maxn,
)
case 2:
mod = model.NewGenAI(
cfg.ModelName,
float32(temp)/100, topp, maxn,
temperature, topp, maxn,
)
default:
logrus.Warnln("[aichat] unsupported AI type", cfg.Type)
@ -319,6 +334,16 @@ func init() {
// 添加群聊总结功能
en.OnRegex(`^群聊总结\s?(\d*)$`, ensureconfig, zero.OnlyGroup, zero.AdminPermission).SetBlock(true).Limit(limit.LimitByGroup).Handle(func(ctx *zero.Ctx) {
ctx.SendChain(message.Text("少女思考中..."))
gid := ctx.Event.GroupID
if gid == 0 {
gid = -ctx.Event.UserID
}
c, ok := ctx.State["manager"].(*ctrl.Control[*zero.Ctx])
if !ok {
return
}
rate := c.GetData(gid)
temp := (rate >> 8) & 0xff
p, _ := strconv.ParseInt(ctx.State["regex_matched"].([]string)[1], 10, 64)
if p > 1000 {
p = 1000
@ -326,10 +351,9 @@ func init() {
if p == 0 {
p = 200
}
gid := ctx.Event.GroupID
group := ctx.GetGroupInfo(gid, false)
if group.MemberCount == 0 {
ctx.SendChain(message.Text(zero.BotConfig.NickName[0], "未加入", group.Name, "(", gid, "),无法获取摘要"))
ctx.SendChain(message.Text(zero.BotConfig.NickName[0], "未加入", group.Name, "(", gid, "),无法获取总结"))
return
}
@ -350,8 +374,13 @@ func init() {
return
}
// 调用大模型API进行摘要
summary, err := summarizeMessages(messages)
// 构造总结请求提示
summaryPrompt := "请总结这个群聊内容,要求按发言顺序梳理,明确标注每个发言者的昵称,并完整呈现其核心观点、提出的问题、发表的看法或做出的回应,确保不遗漏关键信息,且能体现成员间的对话逻辑和互动关系:\n" +
strings.Join(messages, "\n")
// 调用大模型API进行总结
summary, err := llmchat(summaryPrompt, temp)
if err != nil {
ctx.SendChain(message.Text("ERROR: ", err))
return
@ -367,13 +396,108 @@ func init() {
b.WriteString(" 条消息总结:\n\n")
b.WriteString(summary)
// 分割总结内容为多段
parts := strings.Split(b.String(), "\n\n")
msg := make(message.Message, 0, len(parts))
for _, part := range parts {
if part != "" {
msg = append(msg, ctxext.FakeSenderForwardNode(ctx, message.Text(part)))
// 分割总结内容为多段按1000字符长度切割
summaryText := b.String()
msg := make(message.Message, 0)
for len(summaryText) > 0 {
if len(summaryText) <= 1000 {
msg = append(msg, ctxext.FakeSenderForwardNode(ctx, message.Text(summaryText)))
break
}
// 查找1000字符内的最后一个换行符尽量在换行处分割
chunk := summaryText[:1000]
lastNewline := strings.LastIndex(chunk, "\n")
if lastNewline > 0 {
chunk = summaryText[:lastNewline+1]
}
msg = append(msg, ctxext.FakeSenderForwardNode(ctx, message.Text(chunk)))
summaryText = summaryText[len(chunk):]
}
if len(msg) > 0 {
ctx.Send(msg)
}
})
// 添加 /gpt 命令处理(同时支持回复消息和直接使用)
en.OnKeyword("/gpt", ensureconfig).SetBlock(true).Handle(func(ctx *zero.Ctx) {
gid := ctx.Event.GroupID
if gid == 0 {
gid = -ctx.Event.UserID
}
c, ok := ctx.State["manager"].(*ctrl.Control[*zero.Ctx])
if !ok {
return
}
rate := c.GetData(gid)
temp := (rate >> 8) & 0xff
text := ctx.MessageString()
var query string
var replyContent string
// 检查是否是回复消息 (使用MessageElement检查而不是CQ码)
for _, elem := range ctx.Event.Message {
if elem.Type == "reply" {
// 提取被回复的消息ID
replyIDStr := elem.Data["id"]
replyID, err := strconv.ParseInt(replyIDStr, 10, 64)
if err == nil {
// 获取被回复的消息内容
replyMsg := ctx.GetMessage(replyID)
if replyMsg.Elements != nil {
replyContent = replyMsg.Elements.ExtractPlainText()
}
}
break // 找到回复元素后退出循环
}
}
// 提取 /gpt 后面的内容
parts := strings.SplitN(text, "/gpt", 2)
var gContent string
if len(parts) > 1 {
gContent = strings.TrimSpace(parts[1])
}
// 组合内容:优先使用回复内容,如果同时有/gpt内容则拼接
switch {
case replyContent != "" && gContent != "":
query = replyContent + "\n" + gContent
case replyContent != "":
query = replyContent
case gContent != "":
query = gContent
default:
return
}
// 调用大模型API进行聊天
reply, err := llmchat(query, temp)
if err != nil {
ctx.SendChain(message.Text("ERROR: ", err))
return
}
// 分割总结内容为多段按1000字符长度切割
msg := make(message.Message, 0)
for len(reply) > 0 {
if len(reply) <= 1000 {
msg = append(msg, ctxext.FakeSenderForwardNode(ctx, message.Text(reply)))
break
}
// 查找1000字符内的最后一个换行符尽量在换行处分割
chunk := reply[:1000]
lastNewline := strings.LastIndex(chunk, "\n")
if lastNewline > 0 {
chunk = reply[:lastNewline+1]
}
msg = append(msg, ctxext.FakeSenderForwardNode(ctx, message.Text(chunk)))
reply = reply[len(chunk):]
}
if len(msg) > 0 {
ctx.Send(msg)
@ -381,20 +505,34 @@ func init() {
})
}
// summarizeMessages 调用大模型API进行消息摘要
func summarizeMessages(messages []string) (string, error) {
// 使用现有的AI配置进行摘要
// llmchat 调用大模型API包装
func llmchat(prompt string, temp int64) (string, error) {
temperature, topp, maxn := getModelParams(temp) // 使用默认温度70
x := deepinfra.NewAPI(cfg.API, cfg.Key)
mod := model.NewOpenAI(
cfg.ModelName, cfg.Separator,
float32(70)/100, 0.9, 4096,
)
var mod model.Protocol
switch cfg.Type {
case 0:
mod = model.NewOpenAI(
cfg.ModelName, cfg.Separator,
temperature, topp, maxn,
)
case 1:
mod = model.NewOLLaMA(
cfg.ModelName, cfg.Separator,
temperature, topp, maxn,
)
case 2:
mod = model.NewGenAI(
cfg.ModelName,
temperature, topp, maxn,
)
default:
logrus.Warnln("[aichat] unsupported AI type", cfg.Type)
return "", errors.New("不支持的AI类型")
}
// 构造摘要请求提示
summaryPrompt := "请总结这个群聊内容,要求按发言顺序梳理,明确标注每个发言者的昵称,并完整呈现其核心观点、提出的问题、发表的看法或做出的回应,确保不遗漏关键信息,且能体现成员间的对话逻辑和互动关系:\n\n" +
strings.Join(messages, "\n---\n")
data, err := x.Request(mod.User(summaryPrompt))
data, err := x.Request(mod.User(prompt))
if err != nil {
return "", err
}