xinyin025/new-api
1
0
Fork 0
mirror of https://github.com/linux-do/new-api.git synced 2025-11-08 23:23:42 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

feat: support stream_options

Browse Source
This commit is contained in:
CalciumIon
2024-07-08 01:27:57 +08:00
parent 20d71711d3
commit b0e234e8f5
11 changed files with 97 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
relay/channel/ollama/adaptor.go
View File

@@ -59,8 +59,10 @@ func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, request
func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage *dto.Usage, err *dto.OpenAIErrorWithStatusCode) {
if info.IsStream {
var responseText string
err, responseText, _ = openai.OpenaiStreamHandler(c, resp, info)
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
err, usage, responseText, _ = openai.OpenaiStreamHandler(c, resp, info)
if usage == nil || usage.TotalTokens == 0 || (usage.PromptTokens+usage.CompletionTokens) == 0 {
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
}
} else {
if info.RelayMode == relayconstant.RelayModeEmbeddings {
err, usage = ollamaEmbeddingHandler(c, resp, info.PromptTokens, info.UpstreamModelName, info.RelayMode)

24
relay/channel/openai/adaptor.go
View File

@@ -7,6 +7,7 @@ import (
"io"
"net/http"
"one-api/common"
"one-api/constant"
"one-api/dto"
"one-api/relay/channel"
"one-api/relay/channel/ai360"
@@ -19,7 +20,8 @@ import (
)
type Adaptor struct {
ChannelType int
ChannelType int
SupportStreamOptions bool
}
func (a *Adaptor) ConvertRerankRequest(c *gin.Context, relayMode int, request dto.RerankRequest) (any, error) {
@@ -31,6 +33,7 @@ func (a *Adaptor) InitRerank(info *relaycommon.RelayInfo, request dto.RerankRequ
func (a *Adaptor) Init(info *relaycommon.RelayInfo, request dto.GeneralOpenAIRequest) {
a.ChannelType = info.ChannelType
a.SupportStreamOptions = info.SupportStreamOptions
}
func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
@@ -78,6 +81,17 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *dto.Gen
if request == nil {
return nil, errors.New("request is nil")
}
// 如果不支持StreamOptions，将StreamOptions设置为nil
if !a.SupportStreamOptions {
request.StreamOptions = nil
} else {
// 如果支持StreamOptions，且请求中没有设置StreamOptions，根据配置文件设置StreamOptions
if constant.ForceStreamOption {
request.StreamOptions = &dto.StreamOptions{
IncludeUsage: true,
}
}
}
return request, nil
}
@@ -89,9 +103,11 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycom
if info.IsStream {
var responseText string
var toolCount int
err, responseText, toolCount = OpenaiStreamHandler(c, resp, info)
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
usage.CompletionTokens += toolCount * 7
err, usage, responseText, toolCount = OpenaiStreamHandler(c, resp, info)
if usage == nil || usage.TotalTokens == 0 || (usage.PromptTokens+usage.CompletionTokens) == 0 {
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
usage.CompletionTokens += toolCount * 7
}
} else {
err, usage = OpenaiHandler(c, resp, info.PromptTokens, info.UpstreamModelName)
}

24
relay/channel/openai/relay-openai.go
View File

@@ -18,9 +18,10 @@ import (
"time"
)
func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, string, int) {
func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage, string, int) {
//checkSensitive := constant.ShouldCheckCompletionSensitive()
var responseTextBuilder strings.Builder
var usage dto.Usage
toolCount := 0
scanner := bufio.NewScanner(resp.Body)
scanner.Split(func(data []byte, atEOF bool) (advance int, token []byte, err error) {
@@ -62,17 +63,26 @@ func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.
streamItems = append(streamItems, data)
}
}
// 计算token
streamResp := "[" + strings.Join(streamItems, ",") + "]"
switch info.RelayMode {
case relayconstant.RelayModeChatCompletions:
var streamResponses []dto.ChatCompletionsStreamResponseSimple
err := json.Unmarshal(common.StringToByteSlice(streamResp), &streamResponses)
if err != nil {
// 一次性解析失败，逐个解析
common.SysError("error unmarshalling stream response: " + err.Error())
for _, item := range streamItems {
var streamResponse dto.ChatCompletionsStreamResponseSimple
err := json.Unmarshal(common.StringToByteSlice(item), &streamResponse)
if err == nil {
if streamResponse.Usage != nil {
if streamResponse.Usage.TotalTokens != 0 {
usage.PromptTokens += streamResponse.Usage.PromptTokens
usage.CompletionTokens += streamResponse.Usage.CompletionTokens
usage.TotalTokens += streamResponse.Usage.TotalTokens
}
}
for _, choice := range streamResponse.Choices {
responseTextBuilder.WriteString(choice.Delta.GetContentString())
if choice.Delta.ToolCalls != nil {
@@ -89,6 +99,13 @@ func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.
}
} else {
for _, streamResponse := range streamResponses {
if streamResponse.Usage != nil {
if streamResponse.Usage.TotalTokens != 0 {
usage.PromptTokens += streamResponse.Usage.PromptTokens
usage.CompletionTokens += streamResponse.Usage.CompletionTokens
usage.TotalTokens += streamResponse.Usage.TotalTokens
}
}
for _, choice := range streamResponse.Choices {
responseTextBuilder.WriteString(choice.Delta.GetContentString())
if choice.Delta.ToolCalls != nil {
@@ -107,6 +124,7 @@ func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.
var streamResponses []dto.CompletionsStreamResponse
err := json.Unmarshal(common.StringToByteSlice(streamResp), &streamResponses)
if err != nil {
// 一次性解析失败，逐个解析
common.SysError("error unmarshalling stream response: " + err.Error())
for _, item := range streamItems {
var streamResponse dto.CompletionsStreamResponse
@@ -159,10 +177,10 @@ func OpenaiStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.
})
err := resp.Body.Close()
if err != nil {
return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", toolCount
return service.OpenAIErrorWrapperLocal(err, "close_response_body_failed", http.StatusInternalServerError), nil, "", toolCount
}
wg.Wait()
return nil, responseTextBuilder.String(), toolCount
return nil, &usage, responseTextBuilder.String(), toolCount
}
func OpenaiHandler(c *gin.Context, resp *http.Response, promptTokens int, model string) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {

6
relay/channel/perplexity/adaptor.go
View File

@@ -55,8 +55,10 @@ func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, request
func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage *dto.Usage, err *dto.OpenAIErrorWithStatusCode) {
if info.IsStream {
var responseText string
err, responseText, _ = openai.OpenaiStreamHandler(c, resp, info)
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
err, usage, responseText, _ = openai.OpenaiStreamHandler(c, resp, info)
if usage == nil || usage.TotalTokens == 0 || (usage.PromptTokens+usage.CompletionTokens) == 0 {
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
}
} else {
err, usage = openai.OpenaiHandler(c, resp, info.PromptTokens, info.UpstreamModelName)
}

8
relay/channel/zhipu_4v/adaptor.go
View File

@@ -57,9 +57,11 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycom
if info.IsStream {
var responseText string
var toolCount int
err, responseText, toolCount = openai.OpenaiStreamHandler(c, resp, info)
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
usage.CompletionTokens += toolCount * 7
err, usage, responseText, toolCount = openai.OpenaiStreamHandler(c, resp, info)
if usage == nil || usage.TotalTokens == 0 || (usage.PromptTokens+usage.CompletionTokens) == 0 {
usage, _ = service.ResponseText2Usage(responseText, info.UpstreamModelName, info.PromptTokens)
usage.CompletionTokens += toolCount * 7
}
} else {
err, usage = openai.OpenaiHandler(c, resp, info.PromptTokens, info.UpstreamModelName)
}