feat: supper OpenRouter now (close #333, close #340)

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

README.ja.md

@@ -52,7 +52,7 @@ _✨ 標準的な OpenAI API フォーマットを通じてすべての LLM に
 
 > **警告**: この README は ChatGPT によって翻訳されています。翻訳ミスを発見した場合は遠慮なく PR を投稿してください。
 
-> **警告**： 英語版の Docker イメージは `justsong/one-api-ja` です。
+> **警告**： 英語版の Docker イメージは `justsong/one-api-en` です。
 
 > **注**: Docker からプルされた最新のイメージは、`alpha` リリースかもしれません。安定性が必要な場合は、手動でバージョンを指定してください。
 
@@ -89,7 +89,7 @@ _✨ 標準的な OpenAI API フォーマットを通じてすべての LLM に
 
 ## デプロイメント
 ### Docker デプロイメント
-デプロイコマンド: `docker run --name one-api -d --restart always -p 3000:3000 -e TZ=Asia/Shanghai -v /home/ubuntu/data/one-api:/data justsong/one-api-ja`。
+デプロイコマンド: `docker run --name one-api -d --restart always -p 3000:3000 -e TZ=Asia/Shanghai -v /home/ubuntu/data/one-api:/data justsong/one-api-en`。
 
 コマンドを更新する: `docker run --rm -v /var/run/docker.sock:/var/run/docker.sock containrr/watchtower -cR`。
 

README.md

@@ -51,11 +51,13 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
   <a href="https://iamazing.cn/page/reward">赞赏支持</a>
 </p>
 
-> **Note**：本项目为开源项目，使用者必须在遵循 OpenAI 的[使用条款](https://openai.com/policies/terms-of-use)以及**法律法规**的情况下使用，不得用于非法用途。
+> **Note**
+> 本项目为开源项目，使用者必须在遵循 OpenAI 的[使用条款](https://openai.com/policies/terms-of-use)以及**法律法规**的情况下使用，不得用于非法用途。
+> 
+> 根据[《生成式人工智能服务管理暂行办法》](http://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm)的要求，请勿对中国地区公众提供一切未经备案的生成式人工智能服务。
 
-> **Note**：使用 Docker 拉取的最新镜像可能是 `alpha` 版本，如果追求稳定性请手动指定版本。
-
-> **Warning**：从 `v0.3` 版本升级到 `v0.4` 版本需要手动迁移数据库，请手动执行[数据库迁移脚本](./bin/migration_v0.3-v0.4.sql)。
+> **Warning**
+> 使用 Docker 拉取的最新镜像可能是 `alpha` 版本，如果追求稳定性请手动指定版本。
 
 ## 功能
 1. 支持多种大模型：
@@ -66,6 +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

@@ -154,25 +154,27 @@ const (
 )
 
 const (
-	ChannelTypeUnknown   = 0
-	ChannelTypeOpenAI    = 1
-	ChannelTypeAPI2D     = 2
-	ChannelTypeAzure     = 3
-	ChannelTypeCloseAI   = 4
-	ChannelTypeOpenAISB  = 5
-	ChannelTypeOpenAIMax = 6
-	ChannelTypeOhMyGPT   = 7
-	ChannelTypeCustom    = 8
-	ChannelTypeAILS      = 9
-	ChannelTypeAIProxy   = 10
-	ChannelTypePaLM      = 11
-	ChannelTypeAPI2GPT   = 12
-	ChannelTypeAIGC2D    = 13
-	ChannelTypeAnthropic = 14
-	ChannelTypeBaidu     = 15
-	ChannelTypeZhipu     = 16
-	ChannelTypeAli       = 17
-	ChannelTypeXunfei    = 18
+	ChannelTypeUnknown    = 0
+	ChannelTypeOpenAI     = 1
+	ChannelTypeAPI2D      = 2
+	ChannelTypeAzure      = 3
+	ChannelTypeCloseAI    = 4
+	ChannelTypeOpenAISB   = 5
+	ChannelTypeOpenAIMax  = 6
+	ChannelTypeOhMyGPT    = 7
+	ChannelTypeCustom     = 8
+	ChannelTypeAILS       = 9
+	ChannelTypeAIProxy    = 10
+	ChannelTypePaLM       = 11
+	ChannelTypeAPI2GPT    = 12
+	ChannelTypeAIGC2D     = 13
+	ChannelTypeAnthropic  = 14
+	ChannelTypeBaidu      = 15
+	ChannelTypeZhipu      = 16
+	ChannelTypeAli        = 17
+	ChannelTypeXunfei     = 18
+	ChannelType360        = 19
+	ChannelTypeOpenRouter = 20
 )
 
 var ChannelBaseURLs = []string{
@@ -195,4 +197,6 @@ var ChannelBaseURLs = []string{
 	"https://open.bigmodel.cn",       // 16
 	"https://dashscope.aliyuncs.com", // 17
 	"",                               // 18
+	"https://ai.360.cn",              // 19
+	"https://openrouter.ai/api",      // 20
 }

common/model-ratio.go

@@ -1,6 +1,9 @@
 package common
 
-import "encoding/json"
+import (
+	"encoding/json"
+	"strings"
+)
 
 // ModelRatio
 // https://platform.openai.com/docs/models/model-endpoint-compatibility
@@ -10,46 +13,51 @@ import "encoding/json"
 // 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":        0.75,
-	"claude-2":                30,
-	"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 {
@@ -73,3 +81,19 @@ func GetModelRatio(name string) float64 {
 	}
 	return ratio
 }
+
+func GetCompletionRatio(name string) float64 {
+	if strings.HasPrefix(name, "gpt-3.5") {
+		return 1.333333
+	}
+	if strings.HasPrefix(name, "gpt-4") {
+		return 2
+	}
+	if strings.HasPrefix(name, "claude-instant-1") {
+		return 3.38
+	}
+	if strings.HasPrefix(name, "claude-2") {
+		return 2.965517
+	}
+	return 1
+}

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:
@@ -174,7 +178,7 @@ func testAllChannels(notify bool) error {
 				err = errors.New(fmt.Sprintf("响应时间 %.2fs 超过阈值 %.2fs", float64(milliseconds)/1000.0, float64(disableThreshold)/1000.0))
 				disableChannel(channel.Id, channel.Name, err.Error())
 			}
-			if shouldDisableChannel(openaiErr) {
+			if shouldDisableChannel(openaiErr, -1) {
 				disableChannel(channel.Id, channel.Name, err.Error())
 			}
 			channel.UpdateResponseTime(milliseconds)

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-ali.go

@@ -177,9 +177,11 @@ func aliStreamHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithStat
 				common.SysError("error unmarshalling stream response: " + err.Error())
 				return true
 			}
-			usage.PromptTokens += aliResponse.Usage.InputTokens
-			usage.CompletionTokens += aliResponse.Usage.OutputTokens
-			usage.TotalTokens += aliResponse.Usage.InputTokens + aliResponse.Usage.OutputTokens
+			if aliResponse.Usage.OutputTokens != 0 {
+				usage.PromptTokens = aliResponse.Usage.InputTokens
+				usage.CompletionTokens = aliResponse.Usage.OutputTokens
+				usage.TotalTokens = aliResponse.Usage.InputTokens + aliResponse.Usage.OutputTokens
+			}
 			response := streamResponseAli2OpenAI(&aliResponse)
 			response.Choices[0].Delta.Content = strings.TrimPrefix(response.Choices[0].Delta.Content, lastResponseText)
 			lastResponseText = aliResponse.Output.Text

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-baidu.go

@@ -215,9 +215,11 @@ func baiduStreamHandler(c *gin.Context, resp *http.Response) (*OpenAIErrorWithSt
 				common.SysError("error unmarshalling stream response: " + err.Error())
 				return true
 			}
-			usage.PromptTokens += baiduResponse.Usage.PromptTokens
-			usage.CompletionTokens += baiduResponse.Usage.CompletionTokens
-			usage.TotalTokens += baiduResponse.Usage.TotalTokens
+			if baiduResponse.Usage.TotalTokens != 0 {
+				usage.TotalTokens = baiduResponse.Usage.TotalTokens
+				usage.PromptTokens = baiduResponse.Usage.PromptTokens
+				usage.CompletionTokens = baiduResponse.Usage.TotalTokens - baiduResponse.Usage.PromptTokens
+			}
 			response := streamResponseBaidu2OpenAI(&baiduResponse)
 			jsonResponse, err := json.Marshal(response)
 			if err != nil {

controller/relay-text.go

@@ -282,6 +282,10 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 				req.Header.Set("api-key", apiKey)
 			} else {
 				req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
+				if channelType == common.ChannelTypeOpenRouter {
+					req.Header.Set("HTTP-Referer", "https://github.com/songquanpeng/one-api")
+					req.Header.Set("X-Title", "One API")
+				}
 			}
 		case APITypeClaude:
 			req.Header.Set("x-api-key", apiKey)
@@ -315,6 +319,10 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 			return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
 		}
 		isStream = isStream || strings.HasPrefix(resp.Header.Get("Content-Type"), "text/event-stream")
+
+		if resp.StatusCode != http.StatusOK {
+			return relayErrorHandler(resp)
+		}
 	}
 
 	var textResponse TextResponse
@@ -326,14 +334,7 @@ func relayTextHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode {
 		go func() {
 			if consumeQuota {
 				quota := 0
-				completionRatio := 1.0
-				if strings.HasPrefix(textRequest.Model, "gpt-3.5") {
-					completionRatio = 1.333333
-				}
-				if strings.HasPrefix(textRequest.Model, "gpt-4") {
-					completionRatio = 2
-				}
-
+				completionRatio := common.GetCompletionRatio(textRequest.Model)
 				promptTokens = textResponse.Usage.PromptTokens
 				completionTokens = textResponse.Usage.CompletionTokens
 

controller/relay-utils.go

@@ -1,16 +1,38 @@
 package controller
 
 import (
+	"encoding/json"
 	"fmt"
 	"github.com/gin-gonic/gin"
 	"github.com/pkoukk/tiktoken-go"
+	"io"
+	"net/http"
 	"one-api/common"
+	"strconv"
 )
 
 var stopFinishReason = "stop"
 
 var tokenEncoderMap = map[string]*tiktoken.Tiktoken{}
 
+func InitTokenEncoders() {
+	common.SysLog("initializing token encoders")
+	fallbackTokenEncoder, err := tiktoken.EncodingForModel("gpt-3.5-turbo")
+	if err != nil {
+		common.FatalLog(fmt.Sprintf("failed to get fallback token encoder: %s", err.Error()))
+	}
+	for model, _ := range common.ModelRatio {
+		tokenEncoder, err := tiktoken.EncodingForModel(model)
+		if err != nil {
+			common.SysError(fmt.Sprintf("using fallback encoder for model %s", model))
+			tokenEncoderMap[model] = fallbackTokenEncoder
+			continue
+		}
+		tokenEncoderMap[model] = tokenEncoder
+	}
+	common.SysLog("token encoders initialized")
+}
+
 func getTokenEncoder(model string) *tiktoken.Tiktoken {
 	if tokenEncoder, ok := tokenEncoderMap[model]; ok {
 		return tokenEncoder
@@ -95,13 +117,16 @@ func errorWrapper(err error, code string, statusCode int) *OpenAIErrorWithStatus
 	}
 }
 
-func shouldDisableChannel(err *OpenAIError) bool {
+func shouldDisableChannel(err *OpenAIError, statusCode int) bool {
 	if !common.AutomaticDisableChannelEnabled {
 		return false
 	}
 	if err == nil {
 		return false
 	}
+	if statusCode == http.StatusUnauthorized {
+		return true
+	}
 	if err.Type == "insufficient_quota" || err.Code == "invalid_api_key" || err.Code == "account_deactivated" {
 		return true
 	}
@@ -115,3 +140,30 @@ func setEventStreamHeaders(c *gin.Context) {
 	c.Writer.Header().Set("Transfer-Encoding", "chunked")
 	c.Writer.Header().Set("X-Accel-Buffering", "no")
 }
+
+func relayErrorHandler(resp *http.Response) (openAIErrorWithStatusCode *OpenAIErrorWithStatusCode) {
+	openAIErrorWithStatusCode = &OpenAIErrorWithStatusCode{
+		StatusCode: resp.StatusCode,
+		OpenAIError: OpenAIError{
+			Message: fmt.Sprintf("bad response status code %d", resp.StatusCode),
+			Type:    "one_api_error",
+			Code:    "bad_response_status_code",
+			Param:   strconv.Itoa(resp.StatusCode),
+		},
+	}
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		return
+	}
+	var textResponse TextResponse
+	err = json.Unmarshal(responseBody, &textResponse)
+	if err != nil {
+		return
+	}
+	openAIErrorWithStatusCode.OpenAIError = textResponse.Error
+	return
+}

controller/relay-xunfei.go

@@ -75,7 +75,7 @@ type XunfeiChatResponse struct {
 	} `json:"payload"`
 }
 
-func requestOpenAI2Xunfei(request GeneralOpenAIRequest, xunfeiAppId string) *XunfeiChatRequest {
+func requestOpenAI2Xunfei(request GeneralOpenAIRequest, xunfeiAppId string, domain string) *XunfeiChatRequest {
 	messages := make([]XunfeiMessage, 0, len(request.Messages))
 	for _, message := range request.Messages {
 		if message.Role == "system" {
@@ -96,7 +96,7 @@ func requestOpenAI2Xunfei(request GeneralOpenAIRequest, xunfeiAppId string) *Xun
 	}
 	xunfeiRequest := XunfeiChatRequest{}
 	xunfeiRequest.Header.AppId = xunfeiAppId
-	xunfeiRequest.Parameter.Chat.Domain = "general"
+	xunfeiRequest.Parameter.Chat.Domain = domain
 	xunfeiRequest.Parameter.Chat.Temperature = request.Temperature
 	xunfeiRequest.Parameter.Chat.TopK = request.N
 	xunfeiRequest.Parameter.Chat.MaxTokens = request.MaxTokens
@@ -178,15 +178,28 @@ func buildXunfeiAuthUrl(hostUrl string, apiKey, apiSecret string) string {
 
 func xunfeiStreamHandler(c *gin.Context, textRequest GeneralOpenAIRequest, appId string, apiSecret string, apiKey string) (*OpenAIErrorWithStatusCode, *Usage) {
 	var usage Usage
+	query := c.Request.URL.Query()
+	apiVersion := query.Get("api-version")
+	if apiVersion == "" {
+		apiVersion = c.GetString("api_version")
+	}
+	if apiVersion == "" {
+		apiVersion = "v1.1"
+		common.SysLog("api_version not found, use default: " + apiVersion)
+	}
+	domain := "general"
+	if apiVersion == "v2.1" {
+		domain = "generalv2"
+	}
+	hostUrl := fmt.Sprintf("wss://spark-api.xf-yun.com/%s/chat", apiVersion)
 	d := websocket.Dialer{
 		HandshakeTimeout: 5 * time.Second,
 	}
-	hostUrl := "wss://aichat.xf-yun.com/v1/chat"
 	conn, resp, err := d.Dial(buildXunfeiAuthUrl(hostUrl, apiKey, apiSecret), nil)
 	if err != nil || resp.StatusCode != 101 {
 		return errorWrapper(err, "dial_failed", http.StatusInternalServerError), nil
 	}
-	data := requestOpenAI2Xunfei(textRequest, appId)
+	data := requestOpenAI2Xunfei(textRequest, appId, domain)
 	err = conn.WriteJSON(data)
 	if err != nil {
 		return errorWrapper(err, "write_json_failed", http.StatusInternalServerError), 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)
 	}
@@ -185,7 +195,7 @@ func Relay(c *gin.Context) {
 		channelId := c.GetInt("channel_id")
 		common.SysError(fmt.Sprintf("relay error (channel #%d): %s", channelId, err.Message))
 		// https://platform.openai.com/docs/guides/error-codes/api-errors
-		if shouldDisableChannel(&err.OpenAIError) {
+		if shouldDisableChannel(&err.OpenAIError, err.StatusCode) {
 			channelId := c.GetInt("channel_id")
 			channelName := c.GetString("channel_name")
 			disableChannel(channelId, channelName, err.Message)

i18n/en.json

@@ -520,5 +520,8 @@
   "代理": "Proxy",
   "此项可选，用于通过代理站来进行 API 调用，请输入代理站地址，格式为：https://domain.com": "This is optional, used to make API calls through the proxy site, please enter the proxy site address, the format is: https://domain.com",
   "取消密码登录将导致所有未绑定其他登录方式的用户（包括管理员）无法通过密码登录，确认取消？": "Canceling password login will cause all users (including administrators) who have not bound other login methods to be unable to log in via password, confirm cancel?",
-  "按照如下格式输入：": "Enter in the following format:"
+  "按照如下格式输入：": "Enter in the following format:",
+  "模型版本": "Model version",
+  "请输入星火大模型版本，注意是接口地址中的版本号，例如：v2.1": "Please enter the version of the Starfire model, note that it is the version number in the interface address, for example: v2.1",
+  "点击查看": "click to view"
 }

main.go

@@ -77,6 +77,7 @@ func main() {
 		}
 		go controller.AutomaticallyTestChannels(frequency)
 	}
+	controller.InitTokenEncoders()
 
 	// Initialize HTTP server
 	server := gin.Default()

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)
@@ -107,7 +115,7 @@ func Distribute() func(c *gin.Context) {
 		c.Set("model_mapping", channel.ModelMapping)
 		c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))
 		c.Set("base_url", channel.BaseURL)
-		if channel.Type == common.ChannelTypeAzure {
+		if channel.Type == common.ChannelTypeAzure || channel.Type == common.ChannelTypeXunfei {
 			c.Set("api_version", channel.Other)
 		}
 		c.Next()

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/components/LogsTable.js

@@ -43,6 +43,7 @@ function renderType(type) {
 
 const LogsTable = () => {
   const [logs, setLogs] = useState([]);
+  const [showStat, setShowStat] = useState(false);
   const [loading, setLoading] = useState(true);
   const [activePage, setActivePage] = useState(1);
   const [searchKeyword, setSearchKeyword] = useState('');
@@ -92,6 +93,17 @@ const LogsTable = () => {
     }
   };
 
+  const handleEyeClick = async () => {
+    if (!showStat) {
+      if (isAdminUser) {
+        await getLogStat();
+      } else {
+        await getLogSelfStat();
+      }
+    }
+    setShowStat(!showStat);
+  };
+
   const loadLogs = async (startIdx) => {
     let url = '';
     let localStartTimestamp = Date.parse(start_timestamp) / 1000;
@@ -129,13 +141,8 @@ const LogsTable = () => {
 
   const refresh = async () => {
     setLoading(true);
-    setActivePage(1)
+    setActivePage(1);
     await loadLogs(0);
-    if (isAdminUser) {
-      getLogStat().then();
-    } else {
-      getLogSelfStat().then();
-    }
   };
 
   useEffect(() => {
@@ -169,7 +176,7 @@ const LogsTable = () => {
     if (logs.length === 0) return;
     setLoading(true);
     let sortedLogs = [...logs];
-    if (typeof sortedLogs[0][key] === 'string'){
+    if (typeof sortedLogs[0][key] === 'string') {
       sortedLogs.sort((a, b) => {
         return ('' + a[key]).localeCompare(b[key]);
       });
@@ -190,7 +197,12 @@ const LogsTable = () => {
   return (
     <>
       <Segment>
-        <Header as='h3'>使用明细（总消耗额度：{renderQuota(stat.quota)}）</Header>
+        <Header as='h3'>
+          使用明细（总消耗额度：
+          {showStat && renderQuota(stat.quota)}
+          {!showStat && <span onClick={handleEyeClick} style={{ cursor: 'pointer', color: 'gray' }}>点击查看</span>}
+          ）
+        </Header>
         <Form>
           <Form.Group>
             {

web/src/constants/channel.constants.js

@@ -7,7 +7,9 @@ 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: 20, text: '代理：OpenRouter', value: 20, color: 'black' },
   { key: 2, text: '代理：API2D', value: 2, color: 'blue' },
   { key: 5, text: '代理：OpenAI-SB', value: 5, color: 'brown' },
   { key: 7, text: '代理：OhMyGPT', value: 7, color: 'purple' },

web/src/pages/Channel/EditChannel.js

@@ -1,6 +1,6 @@
 import React, { useEffect, useState } from 'react';
 import { Button, Form, Header, Input, Message, Segment } from 'semantic-ui-react';
-import { useParams, useNavigate } from 'react-router-dom';
+import { useNavigate, useParams } from 'react-router-dom';
 import { API, showError, showInfo, showSuccess, verifyJSON } from '../../helpers';
 import { CHANNEL_OPTIONS } from '../../constants';
 
@@ -19,7 +19,7 @@ const EditChannel = () => {
   const handleCancel = () => {
     navigate('/channel');
   };
-  
+
   const originInputs = {
     name: '',
     type: 1,
@@ -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 }));
     }
@@ -163,6 +166,9 @@ const EditChannel = () => {
     if (localInputs.type === 3 && localInputs.other === '') {
       localInputs.other = '2023-06-01-preview';
     }
+    if (localInputs.type === 18 && localInputs.other === '') {
+      localInputs.other = 'v2.1';
+    }
     if (localInputs.model_mapping === '') {
       localInputs.model_mapping = '{}';
     }
@@ -275,6 +281,20 @@ const EditChannel = () => {
               options={groupOptions}
             />
           </Form.Field>
+          {
+            inputs.type === 18 && (
+              <Form.Field>
+                <Form.Input
+                  label='模型版本'
+                  name='other'
+                  placeholder={'请输入星火大模型版本，注意是接口地址中的版本号，例如：v2.1'}
+                  onChange={handleInputChange}
+                  value={inputs.other}
+                  autoComplete='new-password'
+                />
+              </Form.Field>
+            )
+          }
           <Form.Field>
             <Form.Dropdown
               label='模型'