api2gpt-mid/main.go

package main

import (
	"bufio"
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"io/ioutil"
	"log"
	"math/rand"
	"net/http"
	"net/http/httputil"
	"net/url"
	"os"
	"strings"
	"time"

	"github.com/gin-gonic/gin"
	"github.com/pkoukk/tiktoken-go"
	"github.com/redis/go-redis/v9"
)

type Message struct {
	Role    string `json:"role,omitempty"`
	Name    string `json:"name,omitempty"`
	Content string `json:"content,omitempty"`
}

type ChatRequest struct {
	Stream          bool      `json:"stream,omitempty"`
	Model           string    `json:"model,omitempty"`
	PresencePenalty float64   `json:"presence_penalty,omitempty"`
	Temperature     float64   `json:"temperature,omitempty"`
	Messages        []Message `json:"messages,omitempty"`
	Prompt          string    `json:"prompt,omitempty"`
	Input           string    `json:"input,omitempty"`
}

type ChatResponse struct {
	Id      string   `json:"id,omitempty"`
	Object  string   `json:"object,omitempty"`
	Created int64    `json:"created,omitempty"`
	Model   string   `json:"model,omitempty"`
	Usage   Usage    `json:"usage,omitempty"`
	Choices []Choice `json:"choices,omitempty"`
}

type Usage struct {
	PromptTokens     int `json:"prompt_tokens,omitempty"`
	CompletionTokens int `json:"completion_tokens,omitempty"`
	TotalTokens      int `json:"total_tokens,omitempty"`
}

type Choice struct {
	Message      Message `json:"message,omitempty"`
	Delta        Delta   `json:"delta,omitempty"`
	FinishReason string  `json:"finish_reason,omitempty"`
	Index        int     `json:"index,omitempty"`
	Text         string  `json:"text,omitempty"`
}

type Delta struct {
	Content string `json:"content,omitempty"`
}

type CreditSummary struct {
	Object         string  `json:"object"`
	TotalGranted   float64 `json:"total_granted"`
	TotalUsed      float64 `json:"total_used"`
	TotalRemaining float64 `json:"total_remaining"`
}

type ListModelResponse struct {
	Object string  `json:"object"`
	Data   []Model `json:"data"`
}

type Model struct {
	ID         string            `json:"id"`
	Object     string            `json:"object"`
	Created    int               `json:"created"`
	OwnedBy    string            `json:"owned_by"`
	Permission []ModelPermission `json:"permission"`
	Root       string            `json:"root"`
	Parent     any               `json:"parent"`
}

type ModelPermission struct {
	ID                 string `json:"id"`
	Object             string `json:"object"`
	Created            int    `json:"created"`
	AllowCreateEngine  bool   `json:"allow_create_engine"`
	AllowSampling      bool   `json:"allow_sampling"`
	AllowLogprobs      bool   `json:"allow_logprobs"`
	AllowSearchIndices bool   `json:"allow_search_indices"`
	AllowView          bool   `json:"allow_view"`
	AllowFineTuning    bool   `json:"allow_fine_tuning"`
	Organization       string `json:"organization"`
	Group              any    `json:"group"`
	IsBlocking         bool   `json:"is_blocking"`
}

type UserInfo struct {
	UID string `json:"uid"`
	SID string `json:"sid"`
}

type ServerInfo struct {
	ServerAddress string `json:"server_address"`
	AvailableKey  string `json:"available_key"`
}

type ModelInfo struct {
	ModelName       string  `json:"model_name"`
	ModelPrice      float64 `json:"model_price"`
	ModelPrepayment int     `json:"model_prepayment"`
}

var (
	Redis        *redis.Client
	RedisAddress = "localhost:6379"
)

func init() {
	//gin.SetMode(gin.ReleaseMode)
	if v := os.Getenv("REDIS_ADDRESS"); v != "" {
		RedisAddress = v
	}
	log.Printf("loading redis address: %s", RedisAddress)
}

// redis 初始化
func InitRedis() *redis.Client {
	rdb := redis.NewClient(&redis.Options{
		Addr:     RedisAddress,
		Password: "", // no password set
		DB:       0,  // use default DB
		PoolSize: 10,
	})
	result := rdb.Ping(context.Background())
	fmt.Println("redis ping:", result.Val())
	if result.Val() != "PONG" {
		// 连接有问题
		return nil
	}
	return rdb
}

// 计算Messages中的token数量
func numTokensFromMessages(messages []Message, model string) int {
	tkm, err := tiktoken.EncodingForModel(model)
	if err != nil {
		err = fmt.Errorf("getEncoding: %v", err)
		panic(err)
	}
	numTokens := 0
	for _, message := range messages {
		numTokens += len(tkm.Encode(message.Content, nil, nil))
		numTokens += 6
	}
	numTokens += 3
	return numTokens
}

// 计算String中的token数量
func numTokensFromString(msg string, model string) int {
	tkm, err := tiktoken.EncodingForModel(model)
	if err != nil {
		err = fmt.Errorf("getEncoding: %v", err)
		panic(err)
	}
	if model == "text-davinci-003" {
		return len(tkm.Encode(msg, nil, nil)) + 1
	} else {
		return len(tkm.Encode(msg, nil, nil)) + 9
	}
}

func embeddings(c *gin.Context) {
	var prompt_tokens int
	var total_tokens int
	var chatRequest ChatRequest
	if err := c.ShouldBindJSON(&chatRequest); err != nil {
		c.AbortWithStatusJSON(400, gin.H{"error": err.Error()})
		return
	}

	auth := c.Request.Header.Get("Authorization")
	key := strings.Trim(auth, "Bearer ")
	//根据KEY调用用户余额接口，判断是否有足够的余额， 后期可考虑判断max_tokens参数来调整
	serverInfo, err := checkBlance(key, chatRequest.Model)
	if err != nil {
		c.AbortWithStatusJSON(403, gin.H{"error": err.Error()})
		log.Printf("请求出错 KEY: %v Model: %v ERROR: %v", key, chatRequest.Model, err)
		return
	}

	log.Printf("请求的KEY: %v Model: %v", key, chatRequest.Model)

	remote, err := url.Parse(serverInfo.ServerAddress)
	if err != nil {
		panic(err)
	}

	proxy := httputil.NewSingleHostReverseProxy(remote)
	newReqBody, err := json.Marshal(chatRequest)
	if err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	proxy.Director = func(req *http.Request) {
		req.Header = c.Request.Header
		req.Host = remote.Host
		req.URL.Scheme = remote.Scheme
		req.URL.Host = remote.Host
		req.URL.Path = c.Request.URL.Path
		req.ContentLength = int64(len(newReqBody))
		req.Header.Set("Content-Type", "application/json")
		req.Header.Set("Accept-Encoding", "")
		serverKey := serverInfo.AvailableKey
		keyList := strings.Split(serverInfo.AvailableKey, ",")
		if len(keyList) > 1 {
			// 随机数种子
			rand.Seed(time.Now().UnixNano())
			// 从数组中随机选择一个元素
			serverKey = keyList[rand.Intn(len(keyList))]
		}
		req.Header.Set("Authorization", "Bearer "+serverKey)
		req.Body = ioutil.NopCloser(bytes.NewReader(newReqBody))
	}
	sss, err := json.Marshal(chatRequest)
	log.Printf("开始处理返回逻辑 %d", string(sss))

	proxy.ModifyResponse = func(resp *http.Response) error {
		resp.Header.Set("Openai-Organization", "api2gpt")
		var chatResponse ChatResponse
		body, err := ioutil.ReadAll(resp.Body)
		if err != nil {
			return err
		}
		json.Unmarshal(body, &chatResponse)
		prompt_tokens = chatResponse.Usage.PromptTokens
		total_tokens = chatResponse.Usage.TotalTokens
		resp.Body = ioutil.NopCloser(bytes.NewReader(body))
		log.Printf("prompt_tokens:  %v total_tokens: %v", prompt_tokens, total_tokens)
		timestamp := time.Now().Unix()
		timestampID := "emb-" + fmt.Sprintf("%d", timestamp)
		//消费余额
		consumption(key, chatRequest.Model, prompt_tokens, 0, total_tokens, timestampID)
		return nil
	}

	proxy.ServeHTTP(c.Writer, c.Request)

}

func completions(c *gin.Context) {
	var prompt_tokens int
	var completion_tokens int
	var total_tokens int
	var chatRequest ChatRequest
	if err := c.ShouldBindJSON(&chatRequest); err != nil {
		c.AbortWithStatusJSON(400, gin.H{"error": err.Error()})
		return
	}

	auth := c.Request.Header.Get("Authorization")
	key := strings.Trim(auth, "Bearer ")
	//根据KEY调用用户余额接口，判断是否有足够的余额， 后期可考虑判断max_tokens参数来调整
	serverInfo, err := checkBlance(key, chatRequest.Model)
	if err != nil {
		c.AbortWithStatusJSON(403, gin.H{"error": err.Error()})
		log.Printf("请求出错 KEY: %v Model: %v ERROR: %v", key, chatRequest.Model, err)
		return
	}

	log.Printf("请求的KEY: %v Model: %v", key, chatRequest.Model)

	remote, err := url.Parse(serverInfo.ServerAddress)
	if err != nil {
		panic(err)
	}

	proxy := httputil.NewSingleHostReverseProxy(remote)
	newReqBody, err := json.Marshal(chatRequest)
	if err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}
	proxy.Director = func(req *http.Request) {
		req.Header = c.Request.Header
		req.Host = remote.Host
		req.URL.Scheme = remote.Scheme
		req.URL.Host = remote.Host
		req.URL.Path = c.Request.URL.Path
		req.ContentLength = int64(len(newReqBody))
		req.Header.Set("Content-Type", "application/json")
		req.Header.Set("Accept-Encoding", "")
		serverKey := serverInfo.AvailableKey
		keyList := strings.Split(serverInfo.AvailableKey, ",")
		if len(keyList) > 1 {
			// 随机数种子
			rand.Seed(time.Now().UnixNano())
			// 从数组中随机选择一个元素
			serverKey = keyList[rand.Intn(len(keyList))]
		}
		req.Header.Set("Authorization", "Bearer "+serverKey)
		req.Body = ioutil.NopCloser(bytes.NewReader(newReqBody))
	}
	sss, err := json.Marshal(chatRequest)
	log.Printf("开始处理返回逻辑 %d", string(sss))
	if chatRequest.Stream {
		// 流式回应，处理
		proxy.ModifyResponse = func(resp *http.Response) error {
			chatRequestId := ""
			reqContent := ""
			reader := bufio.NewReader(resp.Body)
			headers := resp.Header
			for k, v := range headers {
				c.Writer.Header().Set(k, v[0])
			}
			c.Writer.Header().Set("Openai-Organization", "api2gpt")
			for {
				chunk, err := reader.ReadBytes('\n')
				if err != nil {
					if err == io.EOF {
						break
					}
					log.Printf("httpError1 %v", err.Error())
					break
					//return err
				}
				var chatResponse ChatResponse
				//去除回应中的data:前缀
				var trimStr = strings.Trim(string(chunk), "data: ")
				if trimStr != "\n" {
					json.Unmarshal([]byte(trimStr), &chatResponse)
					//log.Printf("trimStr:" + trimStr)
					if chatResponse.Choices != nil {
						reqContent += chatResponse.Choices[0].Delta.Content
						chatRequestId = chatResponse.Id
					}

					// 写回数据
					_, err = c.Writer.Write([]byte(string(chunk) + "\n"))
					if err != nil {
						log.Printf("httpError2 %v", err.Error())
						return err
					}
					c.Writer.(http.Flusher).Flush()
				}
			}
			if chatRequest.Model == "text-davinci-003" {
				prompt_tokens = numTokensFromString(chatRequest.Prompt, chatRequest.Model)
			} else {
				prompt_tokens = numTokensFromMessages(chatRequest.Messages, chatRequest.Model)
			}
			completion_tokens = numTokensFromString(reqContent, chatRequest.Model)
			total_tokens = prompt_tokens + completion_tokens
			log.Printf("prompt_tokens:  %v completion_tokens:  %v total_tokens: %v", prompt_tokens, completion_tokens, total_tokens)
			//消费余额
			consumption(key, chatRequest.Model, prompt_tokens, completion_tokens, total_tokens, chatRequestId)
			return nil
		}
	} else {
		// 非流式回应，处理
		proxy.ModifyResponse = func(resp *http.Response) error {
			resp.Header.Set("Openai-Organization", "api2gpt")
			var chatResponse ChatResponse
			body, err := ioutil.ReadAll(resp.Body)
			if err != nil {
				return err
			}
			json.Unmarshal(body, &chatResponse)
			prompt_tokens = chatResponse.Usage.PromptTokens
			completion_tokens = chatResponse.Usage.CompletionTokens
			total_tokens = chatResponse.Usage.TotalTokens
			resp.Body = ioutil.NopCloser(bytes.NewReader(body))
			log.Printf("prompt_tokens:  %v completion_tokens:  %v total_tokens: %v", prompt_tokens, completion_tokens, total_tokens)
			//消费余额
			consumption(key, chatRequest.Model, prompt_tokens, completion_tokens, total_tokens, chatResponse.Id)
			return nil
		}
	}

	proxy.ServeHTTP(c.Writer, c.Request)
}

// model查询列表
func handleGetModels(c *gin.Context) {
	// BUGFIX: fix options request, see https://github.com/diemus/azure-openai-proxy/issues/3
	//models := []string{"gpt-4", "gpt-4-0314", "gpt-4-32k", "gpt-4-32k-0314", "gpt-3.5-turbo", "gpt-3.5-turbo-0301", "text-davinci-003", "text-embedding-ada-002"}
	models := []string{"gpt-3.5-turbo", "gpt-3.5-turbo-0301", "text-davinci-003", "text-embedding-ada-002"}
	result := ListModelResponse{
		Object: "list",
	}
	for _, model := range models {
		result.Data = append(result.Data, Model{
			ID:      model,
			Object:  "model",
			Created: 1677649963,
			OwnedBy: "openai",
			Permission: []ModelPermission{
				{
					ID:                 "",
					Object:             "model",
					Created:            1679602087,
					AllowCreateEngine:  true,
					AllowSampling:      true,
					AllowLogprobs:      true,
					AllowSearchIndices: true,
					AllowView:          true,
					AllowFineTuning:    true,
					Organization:       "*",
					Group:              nil,
					IsBlocking:         false,
				},
			},
			Root:   model,
			Parent: nil,
		})
	}
	c.JSON(200, result)
}

// 针对接口预检OPTIONS的处理
func handleOptions(c *gin.Context) {
	// BUGFIX: fix options request, see https://github.com/diemus/azure-openai-proxy/issues/1
	c.Header("Access-Control-Allow-Origin", "*")
	c.Header("Access-Control-Allow-Methods", "POST, GET, OPTIONS, PUT, DELETE")
	c.Header("Access-Control-Allow-Headers", "Content-Type, Authorization")
	c.Status(200)
	return
}

// 余额查询
func balance(c *gin.Context) {
	auth := c.Request.Header.Get("Authorization")
	key := strings.Trim(auth, "Bearer ")
	balance, err := queryBlance(key)
	if err != nil {
		c.JSON(400, gin.H{"error": err.Error()})
	}
	var creditSummary CreditSummary
	creditSummary.Object = "credit_grant"
	creditSummary.TotalGranted = 999999
	creditSummary.TotalUsed = 999999 - balance
	creditSummary.TotalRemaining = balance
	c.JSON(200, creditSummary)
}

// key 检测中间件
func checkKeyMid() gin.HandlerFunc {
	return func(c *gin.Context) {
		auth := c.Request.Header.Get("Authorization")
		log.Printf("auth: %v", auth)
		if auth == "" {
			c.AbortWithStatusJSON(401, gin.H{"code": 40001})
		} else {
			key := strings.Trim(auth, "Bearer ")
			msg, err := checkKeyAndTimeCount(key)
			if err != nil {
				c.AbortWithStatusJSON(msg, gin.H{"code": err.Error()})
			}
		}
		// 执行函数
		c.Next()
	}
}

func main() {

	// 禁用控制台颜色，将日志写入文件时不需要控制台颜色。
	gin.DisableConsoleColor()

	// 记录到文件。
	f, _ := os.Create("gin.log")
	//gin.DefaultWriter = io.MultiWriter(f)
	// 如果需要同时将日志写入文件和控制台，请使用以下代码。
	gin.DefaultWriter = io.MultiWriter(f, os.Stdout)

	log.SetOutput(gin.DefaultWriter)

	r := gin.Default()

	//r.GET("/dashboard/billing/credit_grants", checkKeyMid(), balance)

	r.GET("/v1/models", handleGetModels)
	r.OPTIONS("/v1/*path", handleOptions)
	r.POST("/v1/chat/completions", checkKeyMid(), completions)
	r.POST("/v1/completions", checkKeyMid(), completions)
	r.POST("/v1/embeddings", checkKeyMid(), embeddings)

	r.POST("/mock1", mockBalanceInquiry)
	r.POST("/mock2", mockBalanceConsumption)

	Redis = InitRedis()

	//添加reids测试数据
	// var serverInfo ServerInfo = ServerInfo{
	// 	ServerAddress: "https://gptp.any-door.cn",
	// 	AvailableKey:  "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,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,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",
	// }
	// var serverInfo2 ServerInfo = ServerInfo{
	// 	ServerAddress: "https://azure.any-door.cn",
	// 	AvailableKey:  "6c4d2c65970b40e482e7cd27adb0d119",
	// }
	// serverInfoStr, _ := json.Marshal(&serverInfo)
	// serverInfoStr2, _ := json.Marshal(&serverInfo2)
	// Redis.Set(context.Background(), "server:1", serverInfoStr, 0)
	// Redis.Set(context.Background(), "server:2", serverInfoStr2, 0)

	// var modelInfo ModelInfo = ModelInfo{
	// 	ModelName:       "gpt-3.5-turbo",
	// 	ModelPrice:      0.0001,
	// 	ModelPrepayment: 4000,
	// }
	// modelInfoStr, _ := json.Marshal(&modelInfo)
	// var modelInfo2 ModelInfo = ModelInfo{
	// 	ModelName:       "text-davinci-003",
	// 	ModelPrice:      0.001,
	// 	ModelPrepayment: 4000,
	// }
	// modelInfoStr2, _ := json.Marshal(&modelInfo2)
	// var modelInfo3 ModelInfo = ModelInfo{
	// 	ModelName:       "text-davinci-003",
	// 	ModelPrice:      0.001,
	// 	ModelPrepayment: 4000,
	// }
	// modelInfoStr3, _ := json.Marshal(&modelInfo3)
	// Redis.Set(context.Background(), "model:gpt-3.5-turbo", modelInfoStr, 0)
	// Redis.Set(context.Background(), "model:gpt-3.5-turbo-0301", modelInfoStr, 0)
	// Redis.Set(context.Background(), "model:text-davinci-003", modelInfoStr2, 0)
	// Redis.Set(context.Background(), "model:text-embedding-ada-002", modelInfoStr3, 0)

	// var userInfo UserInfo = UserInfo{
	// 	UID: "1",
	// 	SID: "1",
	// }
	// var userInfo2 UserInfo = UserInfo{
	// 	UID: "2",
	// 	SID: "2",
	// }
	// userInfoStr, _ := json.Marshal(&userInfo)
	// userInfoStr2, _ := json.Marshal(&userInfo2)
	// Redis.Set(context.Background(), "user:8aeb3747-715c-48e8-8b80-aec815949f22", userInfoStr, 0)
	// Redis.Set(context.Background(), "user:AK-7d8ab782-a152-4cc1-9972-568713465c96", userInfoStr2, 0)

	// Redis.IncrByFloat(context.Background(), "user:1:balance", 1000).Result()
	// Redis.IncrByFloat(context.Background(), "user:2:balance", 1000).Result()

	//r.Run("127.0.0.1:8080")
	//docker下使用
	r.Run("0.0.0.0:8080")
}