ai_scheduler/internal/biz/do/prompt.go

169 lines
4.6 KiB
Go

package do
import (
"ai_scheduler/internal/biz/handle"
"ai_scheduler/internal/config"
"ai_scheduler/internal/data/constants"
"ai_scheduler/internal/entitys"
"ai_scheduler/internal/pkg"
"ai_scheduler/internal/pkg/utils_vllm"
"context"
"strings"
"github.com/ollama/ollama/api"
)
type PromptOption interface {
CreatePrompt(ctx context.Context, rec *entitys.Recognize) (mes []api.Message, err error)
}
type WithSys struct {
Config *config.Config
}
func (f *WithSys) CreatePrompt(ctx context.Context, rec *entitys.Recognize) (mes []api.Message, err error) {
var (
prompt = make([]api.Message, 0) // 初始化一个空的api.Message切片
)
// 获取用户内容,如果出错则直接返回错误
content, err := f.getUserContent(ctx, rec)
if err != nil {
return nil, err
}
// 构建提示消息列表,包含系统提示、助手回复和用户内容
mes = append(prompt, api.Message{
Role: "system", // 系统角色
Content: rec.SystemPrompt, // 系统提示内容
}, api.Message{
Role: "assistant", // 助手角色
Content: "### 聊天记录:" + pkg.JsonStringIgonErr(rec.ChatHis), // 助手回复内容
}, api.Message{
Role: "user", // 用户角色
Content: content.String(), // 用户输入内容
})
return
}
func (f *WithSys) getUserContent(ctx context.Context, rec *entitys.Recognize) (content strings.Builder, err error) {
var hasFile bool
if len(rec.UserContent.File) > 0 {
hasFile = true
}
content.WriteString(rec.UserContent.Text)
if hasFile {
content.WriteString("\n")
}
if len(rec.UserContent.Tag) > 0 {
content.WriteString("\n")
content.WriteString("### 工具必须使用:")
content.WriteString(rec.UserContent.Tag)
}
if len(rec.ChatHis.Messages) > 0 {
content.WriteString("### 引用历史聊天记录:\n")
content.WriteString(pkg.JsonStringIgonErr(rec.ChatHis))
}
if hasFile {
content.WriteString("\n")
content.WriteString("### 文件内容:\n")
for _, file := range rec.UserContent.File {
handle.HandleRecognizeFile(file)
// 文件识别
switch file.FileType {
case constants.FileTypeImage:
entitys.ResLog(rec.Ch, "recognize_img_start", "图片识别中...")
var imageContent string
imageContent, err = f.recognizeWithImgVllm(ctx, file)
if err != nil {
return
}
entitys.ResLog(rec.Ch, "recognize_img_end", "图片识别完成,识别内容:"+imageContent)
// 解析结果回写到file
file.FileRec = imageContent
default:
content.WriteString(file.FileRec)
}
}
}
return
}
func (f *WithSys) recognizeWithImgVllm(ctx context.Context, file *entitys.RecognizeFile) (content string, err error) {
if file.FileData == nil || file.FileType != constants.FileTypeImage {
return
}
client, cleanup, err := utils_vllm.NewClient(f.Config)
if err != nil {
return "", err
}
defer cleanup()
outMsg, err := client.RecognizeWithImgBytes(ctx,
f.Config.DefaultPrompt.ImgRecognize.SystemPrompt,
f.Config.DefaultPrompt.ImgRecognize.UserPrompt,
file.FileData,
file.FileRealMime,
)
if err != nil {
return "", err
}
return outMsg.Content, nil
}
type WithDingTalkBot struct {
}
func (f *WithDingTalkBot) CreatePrompt(ctx context.Context, rec *entitys.Recognize) (mes []api.Message, err error) {
var (
prompt = make([]api.Message, 0) // 初始化一个空的api.Message切片
)
// 获取用户内容,如果出错则直接返回错误
content, err := f.getUserContent(ctx, rec)
if err != nil {
return nil, err
}
// 构建提示消息列表,包含系统提示、助手回复和用户内容
mes = append(prompt, api.Message{
Role: "system", // 系统角色
Content: rec.SystemPrompt, // 系统提示内容
}, api.Message{
Role: "assistant", // 助手角色
Content: "### 聊天记录:" + pkg.JsonStringIgonErr(rec.ChatHis), // 助手回复内容
}, api.Message{
Role: "user", // 用户角色
Content: content.String(), // 用户输入内容
})
return
}
func (f *WithDingTalkBot) getUserContent(ctx context.Context, rec *entitys.Recognize) (content strings.Builder, err error) {
var hasFile bool
if rec.UserContent.File != nil && len(rec.UserContent.File) > 0 {
hasFile = true
}
content.WriteString(rec.UserContent.Text)
if hasFile {
content.WriteString("\n")
}
if len(rec.UserContent.Tag) > 0 {
content.WriteString("\n")
content.WriteString("### 工具必须使用:")
content.WriteString(rec.UserContent.Tag)
}
if len(rec.ChatHis.Messages) > 0 {
content.WriteString("### 引用历史聊天记录:\n")
content.WriteString(pkg.JsonStringIgonErr(rec.ChatHis))
}
return
}