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

README.md

@@ -76,7 +76,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [Moonshot AI](https://platform.moonshot.cn/)
    + [x] [百川大模型](https://platform.baichuan-ai.com)
    + [ ] [字节云雀大模型](https://www.volcengine.com/product/ark) (WIP)
-   + [ ] [MINIMAX](https://api.minimax.chat/) (WIP)
+   + [x] [MINIMAX](https://api.minimax.chat/)
 2. 支持配置镜像以及众多[第三方代理服务](https://iamazing.cn/page/openai-api-third-party-services)。
 3. 支持通过**负载均衡**的方式访问多个渠道。
 4. 支持 **stream 模式**，可以通过流式传输实现打字机效果。

common/constants.go

@@ -65,6 +65,7 @@ const (
 	ChannelTypeGemini         = 24
 	ChannelTypeMoonshot       = 25
 	ChannelTypeBaichuan       = 26
+	ChannelTypeMinimax        = 27
 )
 
 var ChannelBaseURLs = []string{
@@ -95,6 +96,7 @@ var ChannelBaseURLs = []string{
 	"https://generativelanguage.googleapis.com", // 24
 	"https://api.moonshot.cn",                   // 25
 	"https://api.baichuan-ai.com",               // 26
+	"https://api.minimax.chat",                  // 27
 }
 
 const (

common/model-ratio.go

@@ -94,14 +94,18 @@ var ModelRatio = map[string]float64{
 	"claude-2.0":              5.51,  // $11.02 / 1M tokens
 	"claude-2.1":              5.51,  // $11.02 / 1M tokens
 	// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7
-	"ERNIE-Bot":                 0.8572,     // ￥0.012 / 1k tokens
-	"ERNIE-Bot-turbo":           0.5715,     // ￥0.008 / 1k tokens
-	"ERNIE-Bot-4":               0.12 * RMB, // ￥0.12 / 1k tokens
-	"ERNIE-Bot-8k":              0.024 * RMB,
-	"Embedding-V1":              0.1429, // ￥0.002 / 1k tokens
-	"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
+	"ERNIE-Bot":         0.8572,     // ￥0.012 / 1k tokens
+	"ERNIE-Bot-turbo":   0.5715,     // ￥0.008 / 1k tokens
+	"ERNIE-Bot-4":       0.12 * RMB, // ￥0.12 / 1k tokens
+	"ERNIE-Bot-8k":      0.024 * RMB,
+	"Embedding-V1":      0.1429, // ￥0.002 / 1k tokens
+	"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
@@ -127,6 +131,23 @@ var ModelRatio = map[string]float64{
 	"moonshot-v1-8k":   0.012 * RMB,
 	"moonshot-v1-32k":  0.024 * RMB,
 	"moonshot-v1-128k": 0.06 * RMB,
+	// https://platform.baichuan-ai.com/price
+	"Baichuan2-Turbo":      0.008 * RMB,
+	"Baichuan2-Turbo-192k": 0.016 * RMB,
+	"Baichuan2-53B":        0.02 * RMB,
+	// https://api.minimax.chat/document/price
+	"abab6-chat":    0.1 * RMB,
+	"abab5.5-chat":  0.015 * RMB,
+	"abab5.5s-chat": 0.005 * RMB,
+}
+
+var DefaultModelRatio map[string]float64
+
+func init() {
+	DefaultModelRatio = make(map[string]float64)
+	for k, v := range ModelRatio {
+		DefaultModelRatio[k] = v
+	}
 }
 
 func ModelRatio2JSONString() string {
@@ -147,6 +168,9 @@ func GetModelRatio(name string) float64 {
 		name = strings.TrimSuffix(name, "-internet")
 	}
 	ratio, ok := ModelRatio[name]
+	if !ok {
+		ratio, ok = DefaultModelRatio[name]
+	}
 	if !ok {
 		logger.SysError("model ratio not found: " + name)
 		return 30

controller/model.go

@@ -5,6 +5,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
 	"github.com/songquanpeng/one-api/relay/channel/baichuan"
+	"github.com/songquanpeng/one-api/relay/channel/minimax"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/helper"
@@ -110,6 +111,17 @@ func init() {
 			Parent:     nil,
 		})
 	}
+	for _, modelName := range minimax.ModelList {
+		openAIModels = append(openAIModels, OpenAIModels{
+			Id:         modelName,
+			Object:     "model",
+			Created:    1626777600,
+			OwnedBy:    "minimax",
+			Permission: permission,
+			Root:       modelName,
+			Parent:     nil,
+		})
+	}
 	openAIModelsMap = make(map[string]OpenAIModels)
 	for _, model := range openAIModels {
 		openAIModelsMap[model.Id] = model

relay/channel/minimax/constants.go

@@ -0,0 +1,7 @@
+package minimax
+
+var ModelList = []string{
+	"abab5.5s-chat",
+	"abab5.5-chat",
+	"abab6-chat",
+}

relay/channel/minimax/main.go

@@ -0,0 +1,14 @@
+package minimax
+
+import (
+	"fmt"
+	"github.com/songquanpeng/one-api/relay/constant"
+	"github.com/songquanpeng/one-api/relay/util"
+)
+
+func GetRequestURL(meta *util.RelayMeta) (string, error) {
+	if meta.Mode == constant.RelayModeChatCompletions {
+		return fmt.Sprintf("%s/v1/text/chatcompletion_v2", meta.BaseURL), nil
+	}
+	return "", fmt.Errorf("unsupported relay mode %d for minimax", meta.Mode)
+}

relay/channel/openai/adaptor.go

@@ -8,6 +8,7 @@ import (
 	"github.com/songquanpeng/one-api/relay/channel"
 	"github.com/songquanpeng/one-api/relay/channel/ai360"
 	"github.com/songquanpeng/one-api/relay/channel/baichuan"
+	"github.com/songquanpeng/one-api/relay/channel/minimax"
 	"github.com/songquanpeng/one-api/relay/channel/moonshot"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/util"
@@ -25,7 +26,8 @@ func (a *Adaptor) Init(meta *util.RelayMeta) {
 }
 
 func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
-	if meta.ChannelType == common.ChannelTypeAzure {
+	switch meta.ChannelType {
+	case common.ChannelTypeAzure:
 		// https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
 		requestURL := strings.Split(meta.RequestURLPath, "?")[0]
 		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, meta.APIVersion)
@@ -39,8 +41,11 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) {
 
 		requestURL = fmt.Sprintf("/openai/deployments/%s/%s", model_, task)
 		return util.GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType), nil
+	case common.ChannelTypeMinimax:
+		return minimax.GetRequestURL(meta)
+	default:
+		return util.GetFullRequestURL(meta.BaseURL, meta.RequestURLPath, meta.ChannelType), nil
 	}
-	return util.GetFullRequestURL(meta.BaseURL, meta.RequestURLPath, meta.ChannelType), nil
 }
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *util.RelayMeta) error {
@@ -71,7 +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)
@@ -87,6 +92,8 @@ func (a *Adaptor) GetModelList() []string {
 		return moonshot.ModelList
 	case common.ChannelTypeBaichuan:
 		return baichuan.ModelList
+	case common.ChannelTypeMinimax:
+		return minimax.ModelList
 	default:
 		return ModelList
 	}
@@ -102,6 +109,8 @@ func (a *Adaptor) GetChannelName() string {
 		return "moonshot"
 	case common.ChannelTypeBaichuan:
 		return "baichuan"
+	case common.ChannelTypeMinimax:
+		return "minimax"
 	default:
 		return "openai"
 	}

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/constants/ChannelConstants.js

@@ -74,7 +74,13 @@ export const CHANNEL_OPTIONS = {
   26: {
     key: 26,
     text: '百川大模型',
-    value: 23,
+    value: 26,
+    color: 'default'
+  },
+  27: {
+    key: 27,
+    text: 'MiniMax',
+    value: 27,
     color: 'default'
   },
   8: {

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",
   },
@@ -157,6 +157,12 @@ const typeConfig = {
     },
     modelGroup: "baichuan",
   },
+  27: {
+    input: {
+      models: ['abab5.5s-chat', 'abab5.5-chat', 'abab6-chat'],
+    },
+    modelGroup: "minimax",
+  },
 };
 
 export { defaultConfig, typeConfig };

web/default/src/constants/channel.constants.js

@@ -12,6 +12,7 @@ export const CHANNEL_OPTIONS = [
   { key: 25, text: 'Moonshot AI', value: 25, color: 'black' },
   { key: 23, text: '腾讯混元', value: 23, color: 'teal' },
   { key: 26, text: '百川大模型', value: 26, color: 'orange' },
+  { key: 27, text: 'MiniMax', value: 27, color: 'red' },
   { key: 8, text: '自定义渠道', value: 8, color: 'pink' },
   { key: 22, text: '知识库：FastGPT', value: 22, color: 'blue' },
   { key: 21, text: '知识库：AI Proxy', value: 21, color: 'purple' },

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 = [
@@ -105,6 +105,9 @@ const EditChannel = () => {
         case 26:
           localModels = ['Baichuan2-Turbo', 'Baichuan2-Turbo-192k', 'Baichuan-Text-Embedding'];
           break;
+        case 27:
+          localModels = ['abab5.5s-chat', 'abab5.5-chat', 'abab6-chat'];
+          break;
       }
       setInputs((inputs) => ({ ...inputs, models: localModels }));
     }