feat: supper whisper now (close #197)

feat: support 360's models (close #331, close #461)

README.md

@@ -68,6 +68,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
    + [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)
    + [x] [智谱 ChatGLM 系列模型](https://bigmodel.cn)
+   + [x] [360 智脑](https://ai.360.cn)
 2. 支持配置镜像以及众多第三方代理服务：
    + [x] [OpenAI-SB](https://openai-sb.com)
    + [x] [API2D](https://api2d.com/r/197971)

common/constants.go

@@ -173,6 +173,7 @@ const (
 	ChannelTypeZhipu     = 16
 	ChannelTypeAli       = 17
 	ChannelTypeXunfei    = 18
+	ChannelType360       = 19
 )
 
 var ChannelBaseURLs = []string{
@@ -195,4 +196,5 @@ var ChannelBaseURLs = []string{
 	"https://open.bigmodel.cn",       // 16
 	"https://dashscope.aliyuncs.com", // 17
 	"",                               // 18
+	"https://ai.360.cn",              // 19
 }

common/model-ratio.go

@@ -13,46 +13,51 @@ import (
 // 1 === $0.002 / 1K tokens
 // 1 === ￥0.014 / 1k tokens
 var ModelRatio = map[string]float64{
-	"gpt-4":                   15,
-	"gpt-4-0314":              15,
-	"gpt-4-0613":              15,
-	"gpt-4-32k":               30,
-	"gpt-4-32k-0314":          30,
-	"gpt-4-32k-0613":          30,
-	"gpt-3.5-turbo":           0.75, // $0.0015 / 1K tokens
-	"gpt-3.5-turbo-0301":      0.75,
-	"gpt-3.5-turbo-0613":      0.75,
-	"gpt-3.5-turbo-16k":       1.5, // $0.003 / 1K tokens
-	"gpt-3.5-turbo-16k-0613":  1.5,
-	"text-ada-001":            0.2,
-	"text-babbage-001":        0.25,
-	"text-curie-001":          1,
-	"text-davinci-002":        10,
-	"text-davinci-003":        10,
-	"text-davinci-edit-001":   10,
-	"code-davinci-edit-001":   10,
-	"whisper-1":               10,
-	"davinci":                 10,
-	"curie":                   10,
-	"babbage":                 10,
-	"ada":                     10,
-	"text-embedding-ada-002":  0.05,
-	"text-search-ada-doc-001": 10,
-	"text-moderation-stable":  0.1,
-	"text-moderation-latest":  0.1,
-	"dall-e":                  8,
-	"claude-instant-1":        0.815,  // $1.63 / 1M tokens
-	"claude-2":                5.51,   // $11.02 / 1M tokens
-	"ERNIE-Bot":               0.8572, // ￥0.012 / 1k tokens
-	"ERNIE-Bot-turbo":         0.5715, // ￥0.008 / 1k tokens
-	"Embedding-V1":            0.1429, // ￥0.002 / 1k tokens
-	"PaLM-2":                  1,
-	"chatglm_pro":             0.7143, // ￥0.01 / 1k tokens
-	"chatglm_std":             0.3572, // ￥0.005 / 1k tokens
-	"chatglm_lite":            0.1429, // ￥0.002 / 1k tokens
-	"qwen-v1":                 0.8572, // TBD: https://help.aliyun.com/document_detail/2399482.html?spm=a2c4g.2399482.0.0.1ad347feilAgag
-	"qwen-plus-v1":            0.5715, // Same as above
-	"SparkDesk":               0.8572, // TBD
+	"gpt-4":                     15,
+	"gpt-4-0314":                15,
+	"gpt-4-0613":                15,
+	"gpt-4-32k":                 30,
+	"gpt-4-32k-0314":            30,
+	"gpt-4-32k-0613":            30,
+	"gpt-3.5-turbo":             0.75, // $0.0015 / 1K tokens
+	"gpt-3.5-turbo-0301":        0.75,
+	"gpt-3.5-turbo-0613":        0.75,
+	"gpt-3.5-turbo-16k":         1.5, // $0.003 / 1K tokens
+	"gpt-3.5-turbo-16k-0613":    1.5,
+	"text-ada-001":              0.2,
+	"text-babbage-001":          0.25,
+	"text-curie-001":            1,
+	"text-davinci-002":          10,
+	"text-davinci-003":          10,
+	"text-davinci-edit-001":     10,
+	"code-davinci-edit-001":     10,
+	"whisper-1":                 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
+	"davinci":                   10,
+	"curie":                     10,
+	"babbage":                   10,
+	"ada":                       10,
+	"text-embedding-ada-002":    0.05,
+	"text-search-ada-doc-001":   10,
+	"text-moderation-stable":    0.1,
+	"text-moderation-latest":    0.1,
+	"dall-e":                    8,
+	"claude-instant-1":          0.815,  // $1.63 / 1M tokens
+	"claude-2":                  5.51,   // $11.02 / 1M tokens
+	"ERNIE-Bot":                 0.8572, // ￥0.012 / 1k tokens
+	"ERNIE-Bot-turbo":           0.5715, // ￥0.008 / 1k tokens
+	"Embedding-V1":              0.1429, // ￥0.002 / 1k tokens
+	"PaLM-2":                    1,
+	"chatglm_pro":               0.7143, // ￥0.01 / 1k tokens
+	"chatglm_std":               0.3572, // ￥0.005 / 1k tokens
+	"chatglm_lite":              0.1429, // ￥0.002 / 1k tokens
+	"qwen-v1":                   0.8572, // TBD: https://help.aliyun.com/document_detail/2399482.html?spm=a2c4g.2399482.0.0.1ad347feilAgag
+	"qwen-plus-v1":              0.5715, // Same as above
+	"SparkDesk":                 0.8572, // TBD
+	"360GPT_S2_V9":              0.8572, // ¥0.012 / 1k tokens
+	"embedding-bert-512-v1":     0.0715, // ¥0.001 / 1k tokens
+	"embedding_s1_v1":           0.0715, // ¥0.001 / 1k tokens
+	"semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
+	"360GPT_S2_V9.4":            0.8572, // ¥0.012 / 1k tokens
 }
 
 func ModelRatio2JSONString() string {

controller/channel-test.go

@@ -24,6 +24,10 @@ func testChannel(channel *model.Channel, request ChatRequest) (error, *OpenAIErr
 		fallthrough
 	case common.ChannelTypeZhipu:
 		fallthrough
+	case common.ChannelTypeAli:
+		fallthrough
+	case common.ChannelType360:
+		fallthrough
 	case common.ChannelTypeXunfei:
 		return errors.New("该渠道类型当前版本不支持测试，请手动测试"), nil
 	case common.ChannelTypeAzure:

controller/model.go

@@ -63,6 +63,15 @@ func init() {
 			Root:       "dall-e",
 			Parent:     nil,
 		},
+		{
+			Id:         "whisper-1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "openai",
+			Permission: permission,
+			Root:       "whisper-1",
+			Parent:     nil,
+		},
 		{
 			Id:         "gpt-3.5-turbo",
 			Object:     "model",
@@ -360,6 +369,51 @@ func init() {
 			Root:       "SparkDesk",
 			Parent:     nil,
 		},
+		{
+			Id:         "360GPT_S2_V9",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "360GPT_S2_V9",
+			Parent:     nil,
+		},
+		{
+			Id:         "embedding-bert-512-v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "embedding-bert-512-v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "embedding_s1_v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "embedding_s1_v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "semantic_similarity_s1_v1",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "semantic_similarity_s1_v1",
+			Parent:     nil,
+		},
+		{
+			Id:         "360GPT_S2_V9.4",
+			Object:     "model",
+			Created:    1677649963,
+			OwnedBy:    "360",
+			Permission: permission,
+			Root:       "360GPT_S2_V9.4",
+			Parent:     nil,
+		},
 	}
 	openAIModelsMap = make(map[string]OpenAIModels)
 	for _, model := range openAIModels {

controller/relay-audio.go

@@ -0,0 +1,147 @@
+package controller
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"one-api/common"
+	"one-api/model"
+
+	"github.com/gin-gonic/gin"
+)
+
+func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
+	audioModel := "whisper-1"
+
+	tokenId := c.GetInt("token_id")
+	channelType := c.GetInt("channel")
+	userId := c.GetInt("id")
+	group := c.GetString("group")
+
+	preConsumedTokens := common.PreConsumedQuota
+	modelRatio := common.GetModelRatio(audioModel)
+	groupRatio := common.GetGroupRatio(group)
+	ratio := modelRatio * groupRatio
+	preConsumedQuota := int(float64(preConsumedTokens) * ratio)
+	userQuota, err := model.CacheGetUserQuota(userId)
+	if err != nil {
+		return errorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError)
+	}
+	err = model.CacheDecreaseUserQuota(userId, preConsumedQuota)
+	if err != nil {
+		return errorWrapper(err, "decrease_user_quota_failed", http.StatusInternalServerError)
+	}
+	if userQuota > 100*preConsumedQuota {
+		// in this case, we do not pre-consume quota
+		// because the user has enough quota
+		preConsumedQuota = 0
+	}
+	if preConsumedQuota > 0 {
+		err := model.PreConsumeTokenQuota(tokenId, preConsumedQuota)
+		if err != nil {
+			return errorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden)
+		}
+	}
+
+	// map model name
+	modelMapping := c.GetString("model_mapping")
+	if modelMapping != "" {
+		modelMap := make(map[string]string)
+		err := json.Unmarshal([]byte(modelMapping), &modelMap)
+		if err != nil {
+			return errorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError)
+		}
+		if modelMap[audioModel] != "" {
+			audioModel = modelMap[audioModel]
+		}
+	}
+
+	baseURL := common.ChannelBaseURLs[channelType]
+	requestURL := c.Request.URL.String()
+
+	if c.GetString("base_url") != "" {
+		baseURL = c.GetString("base_url")
+	}
+
+	fullRequestURL := fmt.Sprintf("%s%s", baseURL, requestURL)
+	requestBody := c.Request.Body
+
+	req, err := http.NewRequest(c.Request.Method, fullRequestURL, requestBody)
+	if err != nil {
+		return errorWrapper(err, "new_request_failed", http.StatusInternalServerError)
+	}
+	req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
+	req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
+	req.Header.Set("Accept", c.Request.Header.Get("Accept"))
+
+	resp, err := httpClient.Do(req)
+	if err != nil {
+		return errorWrapper(err, "do_request_failed", http.StatusInternalServerError)
+	}
+
+	err = req.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
+	}
+	err = c.Request.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
+	}
+	var audioResponse AudioResponse
+
+	defer func() {
+		go func() {
+			quota := countTokenText(audioResponse.Text, audioModel)
+			quotaDelta := quota - preConsumedQuota
+			err := model.PostConsumeTokenQuota(tokenId, quotaDelta)
+			if err != nil {
+				common.SysError("error consuming token remain quota: " + err.Error())
+			}
+			err = model.CacheUpdateUserQuota(userId)
+			if err != nil {
+				common.SysError("error update user quota cache: " + err.Error())
+			}
+			if quota != 0 {
+				tokenName := c.GetString("token_name")
+				logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
+				model.RecordConsumeLog(userId, 0, 0, audioModel, tokenName, quota, logContent)
+				model.UpdateUserUsedQuotaAndRequestCount(userId, quota)
+				channelId := c.GetInt("channel_id")
+				model.UpdateChannelUsedQuota(channelId, quota)
+			}
+		}()
+	}()
+
+	responseBody, err := io.ReadAll(resp.Body)
+
+	if err != nil {
+		return errorWrapper(err, "read_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	err = json.Unmarshal(responseBody, &audioResponse)
+	if err != nil {
+		return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
+	}
+
+	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
+
+	for k, v := range resp.Header {
+		c.Writer.Header().Set(k, v[0])
+	}
+	c.Writer.WriteHeader(resp.StatusCode)
+
+	_, err = io.Copy(c.Writer, resp.Body)
+	if err != nil {
+		return errorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError)
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
+	}
+	return nil
+}

controller/relay.go

@@ -24,6 +24,7 @@ const (
 	RelayModeModerations
 	RelayModeImagesGenerations
 	RelayModeEdits
+	RelayModeAudio
 )
 
 // https://platform.openai.com/docs/api-reference/chat
@@ -40,6 +41,7 @@ type GeneralOpenAIRequest struct {
 	Input       any       `json:"input,omitempty"`
 	Instruction string    `json:"instruction,omitempty"`
 	Size        string    `json:"size,omitempty"`
+	Functions   any       `json:"functions,omitempty"`
 }
 
 type ChatRequest struct {
@@ -62,6 +64,10 @@ type ImageRequest struct {
 	Size   string `json:"size"`
 }
 
+type AudioResponse struct {
+	Text string `json:"text,omitempty"`
+}
+
 type Usage struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
@@ -158,11 +164,15 @@ func Relay(c *gin.Context) {
 		relayMode = RelayModeImagesGenerations
 	} else if strings.HasPrefix(c.Request.URL.Path, "/v1/edits") {
 		relayMode = RelayModeEdits
+	} else if strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+		relayMode = RelayModeAudio
 	}
 	var err *OpenAIErrorWithStatusCode
 	switch relayMode {
 	case RelayModeImagesGenerations:
 		err = relayImageHelper(c, relayMode)
+	case RelayModeAudio:
+		err = relayAudioHelper(c, relayMode)
 	default:
 		err = relayTextHelper(c, relayMode)
 	}

middleware/distributor.go

@@ -58,7 +58,10 @@ func Distribute() func(c *gin.Context) {
 		} else {
 			// Select a channel for the user
 			var modelRequest ModelRequest
-			err := common.UnmarshalBodyReusable(c, &modelRequest)
+			var err error
+			if !strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+				err = common.UnmarshalBodyReusable(c, &modelRequest)
+			}
 			if err != nil {
 				c.JSON(http.StatusBadRequest, gin.H{
 					"error": gin.H{
@@ -84,6 +87,11 @@ func Distribute() func(c *gin.Context) {
 					modelRequest.Model = "dall-e"
 				}
 			}
+			if strings.HasPrefix(c.Request.URL.Path, "/v1/audio") {
+				if modelRequest.Model == "" {
+					modelRequest.Model = "whisper-1"
+				}
+			}
 			channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model)
 			if err != nil {
 				message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", userGroup, modelRequest.Model)

router/relay-router.go

@@ -26,8 +26,8 @@ func SetRelayRouter(router *gin.Engine) {
 		relayV1Router.POST("/images/variations", controller.RelayNotImplemented)
 		relayV1Router.POST("/embeddings", controller.Relay)
 		relayV1Router.POST("/engines/:model/embeddings", controller.Relay)
-		relayV1Router.POST("/audio/transcriptions", controller.RelayNotImplemented)
-		relayV1Router.POST("/audio/translations", controller.RelayNotImplemented)
+		relayV1Router.POST("/audio/transcriptions", controller.Relay)
+		relayV1Router.POST("/audio/translations", controller.Relay)
 		relayV1Router.GET("/files", controller.RelayNotImplemented)
 		relayV1Router.POST("/files", controller.RelayNotImplemented)
 		relayV1Router.DELETE("/files/:id", controller.RelayNotImplemented)

web/src/constants/channel.constants.js

@@ -7,6 +7,7 @@ export const CHANNEL_OPTIONS = [
   { key: 17, text: '阿里通义千问', value: 17, color: 'orange' },
   { key: 18, text: '讯飞星火认知', value: 18, color: 'blue' },
   { key: 16, text: '智谱 ChatGLM', value: 16, color: 'violet' },
+  { key: 19, text: '360 智脑', value: 19, color: 'blue' },
   { key: 8, text: '自定义渠道', value: 8, color: 'pink' },
   { key: 2, text: '代理：API2D', value: 2, color: 'blue' },
   { key: 5, text: '代理：OpenAI-SB', value: 5, color: 'brown' },

web/src/pages/Channel/EditChannel.js

@@ -61,6 +61,9 @@ const EditChannel = () => {
         case 18:
           localModels = ['SparkDesk'];
           break;
+        case 19:
+          localModels = ['360GPT_S2_V9', 'embedding-bert-512-v1', 'embedding_s1_v1', 'semantic_similarity_s1_v1', '360GPT_S2_V9.4']
+          break;
       }
       setInputs((inputs) => ({ ...inputs, models: localModels }));
     }