From e96b173abecef61c02448fb43e2cc07129b5c38a Mon Sep 17 00:00:00 2001 From: GuangxiaoLong <33345078+lgxisbb@users.noreply.github.com> Date: Sun, 24 Mar 2024 21:47:46 +0800 Subject: [PATCH 1/7] =?UTF-8?q?feat:=20=E7=A7=BB=E9=99=A4=20azure=20model?= =?UTF-8?q?=20=E7=9A=84=20TrimSuffix=20(#1193)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- relay/channel/openai/adaptor.go | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) diff --git a/relay/channel/openai/adaptor.go b/relay/channel/openai/adaptor.go index 47594030..1f153c3e 100644 --- a/relay/channel/openai/adaptor.go +++ b/relay/channel/openai/adaptor.go @@ -31,11 +31,8 @@ func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) { task := strings.TrimPrefix(requestURL, "/v1/") model_ := meta.ActualModelName model_ = strings.Replace(model_, ".", "", -1) - // https://github.com/songquanpeng/one-api/issues/67 - model_ = strings.TrimSuffix(model_, "-0301") - model_ = strings.TrimSuffix(model_, "-0314") - model_ = strings.TrimSuffix(model_, "-0613") - + //https://github.com/songquanpeng/one-api/issues/1191 + // {your endpoint}/openai/deployments/{your azure_model}/chat/completions?api-version={api_version} requestURL = fmt.Sprintf("/openai/deployments/%s/%s", model_, task) return util.GetFullRequestURL(meta.BaseURL, requestURL, meta.ChannelType), nil case common.ChannelTypeMinimax: From c243cd553537bf0a544242a52708aaab48e34997 Mon Sep 17 00:00:00 2001 From: xietong Date: Sun, 24 Mar 2024 21:51:31 +0800 Subject: [PATCH 2/7] =?UTF-8?q?feat:=20=E6=94=AF=E6=8C=81=20ollama=20?= =?UTF-8?q?=E7=9A=84=20embedding=20=E6=8E=A5=E5=8F=A3=20(#1221)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * 增加ollama的embedding接口 * chore: fix function name --------- Co-authored-by: JustSong --- relay/channel/ollama/adaptor.go | 18 +++++++-- relay/channel/ollama/main.go | 65 +++++++++++++++++++++++++++++++-- relay/channel/ollama/model.go | 10 +++++ 3 files changed, 86 insertions(+), 7 deletions(-) diff --git a/relay/channel/ollama/adaptor.go b/relay/channel/ollama/adaptor.go index 06c66101..e2ae7d2b 100644 --- a/relay/channel/ollama/adaptor.go +++ b/relay/channel/ollama/adaptor.go @@ -3,13 +3,14 @@ package ollama import ( "errors" "fmt" + "io" + "net/http" + "github.com/gin-gonic/gin" "github.com/songquanpeng/one-api/relay/channel" "github.com/songquanpeng/one-api/relay/constant" "github.com/songquanpeng/one-api/relay/model" "github.com/songquanpeng/one-api/relay/util" - "io" - "net/http" ) type Adaptor struct { @@ -22,6 +23,9 @@ func (a *Adaptor) Init(meta *util.RelayMeta) { func (a *Adaptor) GetRequestURL(meta *util.RelayMeta) (string, error) { // https://github.com/ollama/ollama/blob/main/docs/api.md fullRequestURL := fmt.Sprintf("%s/api/chat", meta.BaseURL) + if meta.Mode == constant.RelayModeEmbeddings { + fullRequestURL = fmt.Sprintf("%s/api/embeddings", meta.BaseURL) + } return fullRequestURL, nil } @@ -37,7 +41,8 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G } switch relayMode { case constant.RelayModeEmbeddings: - return nil, errors.New("not supported") + ollamaEmbeddingRequest := ConvertEmbeddingRequest(*request) + return ollamaEmbeddingRequest, nil default: return ConvertRequest(*request), nil } @@ -51,7 +56,12 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *util.Rel if meta.IsStream { err, usage = StreamHandler(c, resp) } else { - err, usage = Handler(c, resp) + switch meta.Mode { + case constant.RelayModeEmbeddings: + err, usage = EmbeddingHandler(c, resp) + default: + err, usage = Handler(c, resp) + } } return } diff --git a/relay/channel/ollama/main.go b/relay/channel/ollama/main.go index 7ec646a3..821a335b 100644 --- a/relay/channel/ollama/main.go +++ b/relay/channel/ollama/main.go @@ -5,6 +5,10 @@ import ( "context" "encoding/json" "fmt" + "io" + "net/http" + "strings" + "github.com/gin-gonic/gin" "github.com/songquanpeng/one-api/common" "github.com/songquanpeng/one-api/common/helper" @@ -12,9 +16,6 @@ import ( "github.com/songquanpeng/one-api/relay/channel/openai" "github.com/songquanpeng/one-api/relay/constant" "github.com/songquanpeng/one-api/relay/model" - "io" - "net/http" - "strings" ) func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest { @@ -139,6 +140,64 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC return nil, &usage } +func ConvertEmbeddingRequest(request model.GeneralOpenAIRequest) *EmbeddingRequest { + return &EmbeddingRequest{ + Model: request.Model, + Prompt: strings.Join(request.ParseInput(), " "), + } +} + +func EmbeddingHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) { + var ollamaResponse EmbeddingResponse + err := json.NewDecoder(resp.Body).Decode(&ollamaResponse) + if err != nil { + return openai.ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil + } + + err = resp.Body.Close() + if err != nil { + return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil + } + + if ollamaResponse.Error != "" { + return &model.ErrorWithStatusCode{ + Error: model.Error{ + Message: ollamaResponse.Error, + Type: "ollama_error", + Param: "", + Code: "ollama_error", + }, + StatusCode: resp.StatusCode, + }, nil + } + + fullTextResponse := embeddingResponseOllama2OpenAI(&ollamaResponse) + jsonResponse, err := json.Marshal(fullTextResponse) + if err != nil { + return openai.ErrorWrapper(err, "marshal_response_body_failed", http.StatusInternalServerError), nil + } + c.Writer.Header().Set("Content-Type", "application/json") + c.Writer.WriteHeader(resp.StatusCode) + _, err = c.Writer.Write(jsonResponse) + return nil, &fullTextResponse.Usage +} + +func embeddingResponseOllama2OpenAI(response *EmbeddingResponse) *openai.EmbeddingResponse { + openAIEmbeddingResponse := openai.EmbeddingResponse{ + Object: "list", + Data: make([]openai.EmbeddingResponseItem, 0, 1), + Model: "text-embedding-v1", + Usage: model.Usage{TotalTokens: 0}, + } + + openAIEmbeddingResponse.Data = append(openAIEmbeddingResponse.Data, openai.EmbeddingResponseItem{ + Object: `embedding`, + Index: 0, + Embedding: response.Embedding, + }) + return &openAIEmbeddingResponse +} + func Handler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) { ctx := context.TODO() var ollamaResponse ChatResponse diff --git a/relay/channel/ollama/model.go b/relay/channel/ollama/model.go index a8ef1ffc..8baf56a0 100644 --- a/relay/channel/ollama/model.go +++ b/relay/channel/ollama/model.go @@ -35,3 +35,13 @@ type ChatResponse struct { EvalDuration int `json:"eval_duration,omitempty"` Error string `json:"error,omitempty"` } + +type EmbeddingRequest struct { + Model string `json:"model"` + Prompt string `json:"prompt"` +} + +type EmbeddingResponse struct { + Error string `json:"error,omitempty"` + Embedding []float64 `json:"embedding,omitempty"` +} From 99f81a267c6d65c6b426cdc5ea7b001b5f4eff53 Mon Sep 17 00:00:00 2001 From: JustSong Date: Sun, 24 Mar 2024 22:14:45 +0800 Subject: [PATCH 3/7] fix: fix xunfei error handling (close #1218) --- relay/channel/xunfei/main.go | 25 ++++++++++++++----------- 1 file changed, 14 insertions(+), 11 deletions(-) diff --git a/relay/channel/xunfei/main.go b/relay/channel/xunfei/main.go index f89aea2b..5e7014cb 100644 --- a/relay/channel/xunfei/main.go +++ b/relay/channel/xunfei/main.go @@ -121,7 +121,7 @@ func StreamHandler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model) dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId) if err != nil { - return openai.ErrorWrapper(err, "make xunfei request err", http.StatusInternalServerError), nil + return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil } common.SetEventStreamHeaders(c) var usage model.Usage @@ -151,7 +151,7 @@ func Handler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId strin domain, authUrl := getXunfeiAuthUrl(c, apiKey, apiSecret, textRequest.Model) dataChan, stopChan, err := xunfeiMakeRequest(textRequest, domain, authUrl, appId) if err != nil { - return openai.ErrorWrapper(err, "make xunfei request err", http.StatusInternalServerError), nil + return openai.ErrorWrapper(err, "xunfei_request_failed", http.StatusInternalServerError), nil } var usage model.Usage var content string @@ -171,11 +171,7 @@ func Handler(c *gin.Context, textRequest model.GeneralOpenAIRequest, appId strin } } if len(xunfeiResponse.Payload.Choices.Text) == 0 { - xunfeiResponse.Payload.Choices.Text = []ChatResponseTextItem{ - { - Content: "", - }, - } + return openai.ErrorWrapper(err, "xunfei_empty_response_detected", http.StatusInternalServerError), nil } xunfeiResponse.Payload.Choices.Text[0].Content = content @@ -202,15 +198,21 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl, if err != nil { return nil, nil, err } + _, msg, err := conn.ReadMessage() + if err != nil { + return nil, nil, err + } dataChan := make(chan ChatResponse) stopChan := make(chan bool) go func() { for { - _, msg, err := conn.ReadMessage() - if err != nil { - logger.SysError("error reading stream response: " + err.Error()) - break + if msg == nil { + _, msg, err = conn.ReadMessage() + if err != nil { + logger.SysError("error reading stream response: " + err.Error()) + break + } } var response ChatResponse err = json.Unmarshal(msg, &response) @@ -218,6 +220,7 @@ func xunfeiMakeRequest(textRequest model.GeneralOpenAIRequest, domain, authUrl, logger.SysError("error unmarshalling stream response: " + err.Error()) break } + msg = nil dataChan <- response if response.Payload.Choices.Status == 2 { err := conn.Close() From 56ddbb842a9d68a30a0228436d39a888f0d8d87b Mon Sep 17 00:00:00 2001 From: JustSong Date: Sun, 24 Mar 2024 22:20:41 +0800 Subject: [PATCH 4/7] fix: return pre-consumed quota when error happened for audio (close #1217) --- relay/controller/audio.go | 31 +++++++++++++++++++------------ 1 file changed, 19 insertions(+), 12 deletions(-) diff --git a/relay/controller/audio.go b/relay/controller/audio.go index e69c47b3..cd118985 100644 --- a/relay/controller/audio.go +++ b/relay/controller/audio.go @@ -83,6 +83,24 @@ func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus return openai.ErrorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden) } } + succeed := false + defer func() { + if succeed { + return + } + if preConsumedQuota > 0 { + // we need to roll back the pre-consumed quota + defer func(ctx context.Context) { + go func() { + // negative means add quota back for token & user + err := model.PostConsumeTokenQuota(tokenId, -preConsumedQuota) + if err != nil { + logger.Error(ctx, fmt.Sprintf("error rollback pre-consumed quota: %s", err.Error())) + } + }() + }(c.Request.Context()) + } + }() // map model name modelMapping := c.GetString("model_mapping") @@ -193,20 +211,9 @@ func RelayAudioHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus resp.Body = io.NopCloser(bytes.NewBuffer(responseBody)) } if resp.StatusCode != http.StatusOK { - if preConsumedQuota > 0 { - // we need to roll back the pre-consumed quota - defer func(ctx context.Context) { - go func() { - // negative means add quota back for token & user - err := model.PostConsumeTokenQuota(tokenId, -preConsumedQuota) - if err != nil { - logger.Error(ctx, fmt.Sprintf("error rollback pre-consumed quota: %s", err.Error())) - } - }() - }(c.Request.Context()) - } return util.RelayErrorHandler(resp) } + succeed = true quotaDelta := quota - preConsumedQuota defer func(ctx context.Context) { go util.PostConsumeQuota(ctx, tokenId, quotaDelta, quota, userId, channelId, modelRatio, groupRatio, audioModel, tokenName) From cdfdeea3b49bd57fb16eb1dc3586d22b5da67320 Mon Sep 17 00:00:00 2001 From: JustSong Date: Sun, 24 Mar 2024 22:24:41 +0800 Subject: [PATCH 5/7] feat: return token when calling post /api/token (close #1208) --- controller/token.go | 1 + 1 file changed, 1 insertion(+) diff --git a/controller/token.go b/controller/token.go index 7f6b4505..949931da 100644 --- a/controller/token.go +++ b/controller/token.go @@ -142,6 +142,7 @@ func AddToken(c *gin.Context) { c.JSON(http.StatusOK, gin.H{ "success": true, "message": "", + "data": cleanToken, }) return } From f76c46d6489f0be68a1813f1c4b2d1e066c22cb3 Mon Sep 17 00:00:00 2001 From: JustSong Date: Sun, 24 Mar 2024 22:50:09 +0800 Subject: [PATCH 6/7] feat: add gemini-1.5-pro (#1211) --- common/model-ratio.go | 12 +++++++++--- relay/channel/gemini/constants.go | 2 +- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/common/model-ratio.go b/common/model-ratio.go index 2c608302..460d4843 100644 --- a/common/model-ratio.go +++ b/common/model-ratio.go @@ -81,9 +81,12 @@ var ModelRatio = map[string]float64{ "bge-large-en": 0.002 * RMB, "bge-large-8k": 0.002 * RMB, // https://ai.google.dev/pricing - "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 + "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 + "gemini-1.0-pro-vision-001": 1, + "gemini-1.0-pro-001": 1, + "gemini-1.5-pro": 1, // https://open.bigmodel.cn/pricing "glm-4": 0.1 * RMB, "glm-4v": 0.1 * RMB, @@ -249,6 +252,9 @@ func GetCompletionRatio(name string) float64 { if strings.HasPrefix(name, "mistral-") { return 3 } + if strings.HasPrefix(name, "gemini-") { + return 3 + } switch name { case "llama2-70b-4096": return 0.8 / 0.7 diff --git a/relay/channel/gemini/constants.go b/relay/channel/gemini/constants.go index e8d3a155..32e7c240 100644 --- a/relay/channel/gemini/constants.go +++ b/relay/channel/gemini/constants.go @@ -3,6 +3,6 @@ package gemini // https://ai.google.dev/models/gemini var ModelList = []string{ - "gemini-pro", "gemini-1.0-pro-001", + "gemini-pro", "gemini-1.0-pro-001", "gemini-1.5-pro", "gemini-pro-vision", "gemini-1.0-pro-vision-001", } From 5b349efff9906d6db9f31644008959250b2c30f9 Mon Sep 17 00:00:00 2001 From: JustSong Date: Sun, 24 Mar 2024 22:57:24 +0800 Subject: [PATCH 7/7] chore: fix berry copy --- web/berry/src/views/Authentication/Auth/Register.js | 2 +- web/berry/src/views/Authentication/AuthForms/AuthRegister.js | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/web/berry/src/views/Authentication/Auth/Register.js b/web/berry/src/views/Authentication/Auth/Register.js index 4489e560..8027649d 100644 --- a/web/berry/src/views/Authentication/Auth/Register.js +++ b/web/berry/src/views/Authentication/Auth/Register.js @@ -51,7 +51,7 @@ const Register = () => { - 已经有帐号了?点击登录 + 已经有帐号了？点击登录 diff --git a/web/berry/src/views/Authentication/AuthForms/AuthRegister.js b/web/berry/src/views/Authentication/AuthForms/AuthRegister.js index c286faad..8d588696 100644 --- a/web/berry/src/views/Authentication/AuthForms/AuthRegister.js +++ b/web/berry/src/views/Authentication/AuthForms/AuthRegister.js @@ -296,7 +296,7 @@ const RegisterForm = ({ ...others }) => {