feat: support chatglm-4 (close #1045, close #952, close #952, close #943)

common/model-ratio.go

@@ -102,6 +102,10 @@ var ModelRatio = map[string]float64{
 	"PaLM-2":            1,
 	"gemini-pro":        1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
 	"gemini-pro-vision": 1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	// https://open.bigmodel.cn/pricing
+	"glm-4":                     0.1 * RMB,
+	"glm-4v":                    0.1 * RMB,
+	"glm-3-turbo":               0.005 * RMB,
 	"chatglm_turbo":             0.3572, // ￥0.005 / 1k tokens
 	"chatglm_pro":               0.7143, // ￥0.01 / 1k tokens
 	"chatglm_std":               0.3572, // ￥0.005 / 1k tokens

relay/channel/openai/adaptor.go

@@ -76,7 +76,7 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
 		var responseText string
-		err, responseText = StreamHandler(c, resp, meta.Mode)
+		err, responseText, _ = StreamHandler(c, resp, meta.Mode)
 		usage = ResponseText2Usage(responseText, meta.ActualModelName, meta.PromptTokens)
 	} else {
 		err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)

relay/channel/openai/main.go

@@ -14,7 +14,7 @@ import (
 	"strings"
 )
 
-func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string) {
+func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
 	responseText := ""
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
@@ -31,6 +31,7 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 	})
 	dataChan := make(chan string)
 	stopChan := make(chan bool)
+	var usage *model.Usage
 	go func() {
 		for scanner.Scan() {
 			data := scanner.Text()
@@ -54,6 +55,9 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 					for _, choice := range streamResponse.Choices {
 						responseText += choice.Delta.Content
 					}
+					if streamResponse.Usage != nil {
+						usage = streamResponse.Usage
+					}
 				case constant.RelayModeCompletions:
 					var streamResponse CompletionsStreamResponse
 					err := json.Unmarshal([]byte(data), &streamResponse)
@@ -86,9 +90,9 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 	})
 	err := resp.Body.Close()
 	if err != nil {
-		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), ""
+		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", nil
 	}
-	return nil, responseText
+	return nil, responseText, usage
 }
 
 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {

relay/channel/openai/model.go

@@ -132,6 +132,7 @@ type ChatCompletionsStreamResponse struct {
 	Created int64                                 `json:"created"`
 	Model   string                                `json:"model"`
 	Choices []ChatCompletionsStreamResponseChoice `json:"choices"`
+	Usage   *model.Usage                          `json:"usage"`
 }
 
 type CompletionsStreamResponse struct {

relay/channel/tencent/main.go

@@ -81,6 +81,7 @@ func responseTencent2OpenAI(response *ChatResponse) *openai.TextResponse {
 
 func streamResponseTencent2OpenAI(TencentResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
 	response := openai.ChatCompletionsStreamResponse{
+		Id:      fmt.Sprintf("chatcmpl-%s", helper.GetUUID()),
 		Object:  "chat.completion.chunk",
 		Created: helper.GetTimestamp(),
 		Model:   "tencent-hunyuan",

relay/channel/zhipu/adaptor.go

@@ -5,20 +5,35 @@ import (
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/relay/channel"
+	"github.com/songquanpeng/one-api/relay/channel/openai"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
 	"io"
 	"net/http"
+	"strings"
 )
 
 type Adaptor struct {
+	APIVersion string
 }
 
 func (a *Adaptor) Init(meta *util.RelayMeta) {
 
 }
 
+func (a *Adaptor) SetVersionByModeName(modelName string) {
+	if strings.HasPrefix(modelName, "glm-") {
+		a.APIVersion = "v4"
+	} else {
+		a.APIVersion = "v3"
+	}
+}
+
 func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
+	a.SetVersionByModeName(meta.ActualModelName)
+	if a.APIVersion == "v4" {
+		return fmt.Sprintf("%s/api/paas/v4/chat/completions", meta.BaseURL), nil
+	}
 	method := "invoke"
 	if meta.IsStream {
 		method = "sse-invoke"
@@ -37,6 +52,13 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
+	if request.TopP >= 1 {
+		request.TopP = 0.99
+	}
+	a.SetVersionByModeName(request.Model)
+	if a.APIVersion == "v4" {
+		return request, nil
+	}
 	return ConvertRequest(*request), nil
 }
 
@@ -44,7 +66,19 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *util.RelayMeta, requestBody io
 	return channel.DoRequestHelper(a, c, meta, requestBody)
 }
 
+func (a *Adaptor) DoResponseV4(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if meta.IsStream {
+		err, _, usage = openai.StreamHandler(c, resp, meta.Mode)
+	} else {
+		err, usage = openai.Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
+	}
+	return
+}
+
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.RelayMeta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	if a.APIVersion == "v4" {
+		return a.DoResponseV4(c, resp, meta)
+	}
 	if meta.IsStream {
 		err, usage = StreamHandler(c, resp)
 	} else {

relay/channel/zhipu/constants.go

@@ -2,4 +2,5 @@ package zhipu
 
 var ModelList = []string{
 	"chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite",
+	"glm-4", "glm-4v", "glm-3-turbo",
 }

web/berry/src/views/Channel/type/Config.js

@@ -67,7 +67,7 @@ const typeConfig = {
   },
   16: {
     input: {
-      models: ["chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite"],
+      models: ["glm-4", "glm-4v", "glm-3-turbo", "chatglm_turbo", "chatglm_pro", "chatglm_std", "chatglm_lite"],
     },
     modelGroup: "zhipu",
   },

web/default/src/pages/Channel/EditChannel.js

@@ -79,7 +79,7 @@ const EditChannel = () => {
           localModels = [...localModels, ...withInternetVersion];
           break;
         case 16:
-          localModels = ['chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
+          localModels = ["glm-4", "glm-4v", "glm-3-turbo",'chatglm_turbo', 'chatglm_pro', 'chatglm_std', 'chatglm_lite'];
           break;
         case 18:
           localModels = [