xinyin025/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-12-26 01:35:58 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/upstream/main'

Browse Source
This commit is contained in:
Laisky.Cai
2024-07-23 01:14:23 +00:00
parent 5aa26d691c c936198ac8
commit 4c96688e7c
22 changed files with 352 additions and 145 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
relay/adaptor.go
View File

@@ -15,6 +15,7 @@ import (
"github.com/songquanpeng/one-api/relay/adaptor/ollama"
"github.com/songquanpeng/one-api/relay/adaptor/openai"
"github.com/songquanpeng/one-api/relay/adaptor/palm"
"github.com/songquanpeng/one-api/relay/adaptor/proxy"
"github.com/songquanpeng/one-api/relay/adaptor/tencent"
"github.com/songquanpeng/one-api/relay/adaptor/vertexai"
"github.com/songquanpeng/one-api/relay/adaptor/xunfei"
@@ -58,6 +59,8 @@ func GetAdaptor(apiType int) adaptor.Adaptor {
return &deepl.Adaptor{}
case apitype.VertexAI:
return &vertexai.Adaptor{}
case apitype.Proxy:
return &proxy.Adaptor{}
}
return nil

6
relay/adaptor/doubao/main.go
View File

@@ -7,8 +7,12 @@ import (
)
func GetRequestURL(meta *meta.Meta) (string, error) {
if meta.Mode == relaymode.ChatCompletions {
switch meta.Mode {
case relaymode.ChatCompletions:
return fmt.Sprintf("%s/api/v3/chat/completions", meta.BaseURL), nil
case relaymode.Embeddings:
return fmt.Sprintf("%s/api/v3/embeddings", meta.BaseURL), nil
default:
}
return "", fmt.Errorf("unsupported relay mode %d for doubao", meta.Mode)
}

1
relay/adaptor/openai/constants.go
View File

@@ -8,6 +8,7 @@ var ModelList = []string{
"gpt-4-32k", "gpt-4-32k-0314", "gpt-4-32k-0613",
"gpt-4-turbo-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09",
"gpt-4o", "gpt-4o-2024-05-13",
"gpt-4o-mini", "gpt-4o-mini-2024-07-18",
"gpt-4-vision-preview",
"text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large",
"text-curie-001", "text-babbage-001", "text-ada-001", "text-davinci-002", "text-davinci-003",

14
relay/adaptor/openai/token.go
View File

@@ -110,7 +110,7 @@ func CountTokenMessages(messages []model.Message, model string) int {
if imageUrl["detail"] != nil {
detail = imageUrl["detail"].(string)
}
imageTokens, err := countImageTokens(url, detail)
imageTokens, err := countImageTokens(url, detail, model)
if err != nil {
logger.SysError("error counting image tokens: " + err.Error())
} else {
@@ -181,11 +181,15 @@ const (
lowDetailCost = 85
highDetailCostPerTile = 170
additionalCost = 85
// gpt-4o-mini cost higher than other model
gpt4oMiniLowDetailCost = 2833
gpt4oMiniHighDetailCost = 5667
gpt4oMiniAdditionalCost = 2833
)
// https://platform.openai.com/docs/guides/vision/calculating-costs
// https://github.com/openai/openai-cookbook/blob/05e3f9be4c7a2ae7ecf029a7c32065b024730ebe/examples/How_to_count_tokens_with_tiktoken.ipynb
func countImageTokens(url string, detail string) (_ int, err error) {
func countImageTokens(url string, detail string, model string) (_ int, err error) {
var fetchSize = true
var width, height int
// Reference: https://platform.openai.com/docs/guides/vision/low-or-high-fidelity-image-understanding
@@ -219,6 +223,9 @@ func countImageTokens(url string, detail string) (_ int, err error) {
}
switch detail {
case "low":
if strings.HasPrefix(model, "gpt-4o-mini") {
return gpt4oMiniLowDetailCost, nil
}
return lowDetailCost, nil
case "high":
if fetchSize {
@@ -238,6 +245,9 @@ func countImageTokens(url string, detail string) (_ int, err error) {
height = int(float64(height) * ratio)
}
numSquares := int(math.Ceil(float64(width)/512) * math.Ceil(float64(height)/512))
if strings.HasPrefix(model, "gpt-4o-mini") {
return numSquares*gpt4oMiniHighDetailCost + gpt4oMiniAdditionalCost, nil
}
result := numSquares*highDetailCostPerTile + additionalCost
return result, nil
default:

89
relay/adaptor/proxy/adaptor.go Normal file
View File

@@ -0,0 +1,89 @@
package proxy
import (
"fmt"
"io"
"net/http"
"strings"
"github.com/gin-gonic/gin"
"github.com/pkg/errors"
"github.com/songquanpeng/one-api/relay/adaptor"
channelhelper "github.com/songquanpeng/one-api/relay/adaptor"
"github.com/songquanpeng/one-api/relay/meta"
"github.com/songquanpeng/one-api/relay/model"
relaymodel "github.com/songquanpeng/one-api/relay/model"
)
var _ adaptor.Adaptor = new(Adaptor)
const channelName = "proxy"
type Adaptor struct{}
func (a *Adaptor) Init(meta *meta.Meta) {
}
func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error) {
return nil, errors.New("notimplement")
}
func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
for k, v := range resp.Header {
for _, vv := range v {
c.Writer.Header().Set(k, vv)
}
}
c.Writer.WriteHeader(resp.StatusCode)
if _, gerr := io.Copy(c.Writer, resp.Body); gerr != nil {
return nil, &relaymodel.ErrorWithStatusCode{
StatusCode: http.StatusInternalServerError,
Error: relaymodel.Error{
Message: gerr.Error(),
},
}
}
return nil, nil
}
func (a *Adaptor) GetModelList() (models []string) {
return nil
}
func (a *Adaptor) GetChannelName() string {
return channelName
}
// GetRequestURL remove static prefix, and return the real request url to the upstream service
func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
prefix := fmt.Sprintf("/v1/oneapi/proxy/%d", meta.ChannelId)
return meta.BaseURL + strings.TrimPrefix(meta.RequestURLPath, prefix), nil
}
func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
for k, v := range c.Request.Header {
req.Header.Set(k, v[0])
}
// remove unnecessary headers
req.Header.Del("Host")
req.Header.Del("Content-Length")
req.Header.Del("Accept-Encoding")
req.Header.Del("Connection")
// set authorization header
req.Header.Set("Authorization", meta.APIKey)
return nil
}
func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
return nil, errors.Errorf("not implement")
}
func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error) {
return channelhelper.DoRequestHelper(a, c, meta, requestBody)
}

1
relay/apitype/define.go
View File

@@ -18,6 +18,7 @@ const (
Cloudflare
DeepL
VertexAI
Proxy
Dummy // this one is only for count, do not add any channel after this
)

23
relay/billing/ratio/model.go
View File

@@ -28,15 +28,17 @@ var ModelRatio = map[string]float64{
"gpt-4-32k": 30,
"gpt-4-32k-0314": 30,
"gpt-4-32k-0613": 30,
"gpt-4-1106-preview": 5, // $0.01 / 1K tokens
"gpt-4-0125-preview": 5, // $0.01 / 1K tokens
"gpt-4-turbo-preview": 5, // $0.01 / 1K tokens
"gpt-4-turbo": 5, // $0.01 / 1K tokens
"gpt-4-turbo-2024-04-09": 5, // $0.01 / 1K tokens
"gpt-4o": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-05-13": 2.5, // $0.005 / 1K tokens
"gpt-4-vision-preview": 5, // $0.01 / 1K tokens
"gpt-3.5-turbo": 0.25, // $0.0005 / 1K tokens
"gpt-4-1106-preview": 5, // $0.01 / 1K tokens
"gpt-4-0125-preview": 5, // $0.01 / 1K tokens
"gpt-4-turbo-preview": 5, // $0.01 / 1K tokens
"gpt-4-turbo": 5, // $0.01 / 1K tokens
"gpt-4-turbo-2024-04-09": 5, // $0.01 / 1K tokens
"gpt-4o": 2.5, // $0.005 / 1K tokens
"gpt-4o-2024-05-13": 2.5, // $0.005 / 1K tokens
"gpt-4o-mini": 0.075, // $0.00015 / 1K tokens
"gpt-4o-mini-2024-07-18": 0.075, // $0.00015 / 1K tokens
"gpt-4-vision-preview": 5, // $0.01 / 1K tokens
"gpt-3.5-turbo": 0.25, // $0.0005 / 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
@@ -308,6 +310,9 @@ func GetCompletionRatio(name string, channelType int) float64 {
return 4.0 / 3.0
}
if strings.HasPrefix(name, "gpt-4") {
if strings.HasPrefix(name, "gpt-4o-mini") {
return 4
}
if strings.HasPrefix(name, "gpt-4-turbo") ||
strings.HasPrefix(name, "gpt-4o") ||
strings.HasSuffix(name, "preview") {

1
relay/channeltype/define.go
View File

@@ -44,5 +44,6 @@ const (
Doubao
Novita
VertextAI
Proxy
Dummy
)

2
relay/channeltype/helper.go
View File

@@ -37,6 +37,8 @@ func ToAPIType(channelType int) int {
apiType = apitype.DeepL
case VertextAI:
apiType = apitype.VertexAI
case Proxy:
apiType = apitype.Proxy
}
return apiType

1
relay/channeltype/url.go
View File

@@ -44,6 +44,7 @@ var ChannelBaseURLs = []string{
"https://ark.cn-beijing.volces.com", // 40
"https://api.novita.ai/v3/openai", // 41
"", // 42
"", // 43
}
func init() {

41
relay/controller/proxy.go Normal file
View File

@@ -0,0 +1,41 @@
// Package controller is a package for handling the relay controller
package controller
import (
"fmt"
"net/http"
"github.com/gin-gonic/gin"
"github.com/songquanpeng/one-api/common/logger"
"github.com/songquanpeng/one-api/relay"
"github.com/songquanpeng/one-api/relay/adaptor/openai"
"github.com/songquanpeng/one-api/relay/meta"
relaymodel "github.com/songquanpeng/one-api/relay/model"
)
// RelayProxyHelper is a helper function to proxy the request to the upstream service
func RelayProxyHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatusCode {
ctx := c.Request.Context()
meta := meta.GetByContext(c)
adaptor := relay.GetAdaptor(meta.APIType)
if adaptor == nil {
return openai.ErrorWrapper(fmt.Errorf("invalid api type: %d", meta.APIType), "invalid_api_type", http.StatusBadRequest)
}
adaptor.Init(meta)
resp, err := adaptor.DoRequest(c, meta, c.Request.Body)
if err != nil {
logger.Errorf(ctx, "DoRequest failed: %s", err.Error())
return openai.ErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
}
// do response
_, respErr := adaptor.DoResponse(c, resp, meta)
if respErr != nil {
logger.Errorf(ctx, "respErr is not nil: %+v", respErr)
return respErr
}
return nil
}

54
relay/controller/text.go
View File

@@ -12,6 +12,7 @@ import (
"github.com/songquanpeng/one-api/common/logger"
"github.com/songquanpeng/one-api/model"
"github.com/songquanpeng/one-api/relay"
"github.com/songquanpeng/one-api/relay/adaptor"
"github.com/songquanpeng/one-api/relay/adaptor/openai"
"github.com/songquanpeng/one-api/relay/apitype"
"github.com/songquanpeng/one-api/relay/billing"
@@ -33,9 +34,8 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
meta.IsStream = textRequest.Stream
// map model name
var isModelMapped bool
meta.OriginModelName = textRequest.Model
textRequest.Model, isModelMapped = getMappedModelName(textRequest.Model, meta.ModelMapping)
textRequest.Model, _ = getMappedModelName(textRequest.Model, meta.ModelMapping)
meta.ActualModelName = textRequest.Model
// get model ratio & group ratio
modelRatio := billingratio.GetModelRatio(textRequest.Model, meta.ChannelType)
@@ -59,30 +59,9 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
adaptor.Init(meta)
// get request body
var requestBody io.Reader
if meta.APIType == apitype.OpenAI {
// no need to convert request for openai
shouldResetRequestBody := isModelMapped || meta.ChannelType == channeltype.Baichuan // frequency_penalty 0 is not acceptable for baichuan
if shouldResetRequestBody {
jsonStr, err := json.Marshal(textRequest)
if err != nil {
return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
}
requestBody = bytes.NewBuffer(jsonStr)
} else {
requestBody = c.Request.Body
}
} else {
convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
if err != nil {
return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
}
jsonData, err := json.Marshal(convertedRequest)
if err != nil {
return openai.ErrorWrapper(err, "json_marshal_failed", http.StatusInternalServerError)
}
logger.Debugf(ctx, "converted request: \n%s", string(jsonData))
requestBody = bytes.NewBuffer(jsonData)
requestBody, err := getRequestBody(c, meta, textRequest, adaptor)
if err != nil {
return openai.ErrorWrapper(err, "convert_request_failed", http.StatusInternalServerError)
}
// for debug
@@ -123,3 +102,26 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
return nil
}
func getRequestBody(c *gin.Context, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, adaptor adaptor.Adaptor) (io.Reader, error) {
if meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
// no need to convert request for openai
return c.Request.Body, nil
}
// get request body
var requestBody io.Reader
convertedRequest, err := adaptor.ConvertRequest(c, meta.Mode, textRequest)
if err != nil {
logger.Debugf(c.Request.Context(), "converted request failed: %s\n", err.Error())
return nil, err
}
jsonData, err := json.Marshal(convertedRequest)
if err != nil {
logger.Debugf(c.Request.Context(), "converted request json_marshal_failed: %s\n", err.Error())
return nil, err
}
logger.Debugf(c.Request.Context(), "converted request: \n%s", string(jsonData))
requestBody = bytes.NewBuffer(jsonData)
return requestBody, nil
}

11
relay/meta/relay_meta.go
View File

@@ -19,11 +19,12 @@ type Meta struct {
UserId int
Group string
ModelMapping map[string]string
BaseURL string
APIKey string
APIType int
Config model.ChannelConfig
IsStream bool
// BaseURL is the proxy url set in the channel config
BaseURL string
APIKey string
APIType int
Config model.ChannelConfig
IsStream bool
// OriginModelName is the model name from the raw user request
OriginModelName string
// ActualModelName is the model name after mapping

2
relay/relaymode/define.go
View File

@@ -11,5 +11,7 @@ const (
AudioSpeech
AudioTranscription
AudioTranslation
// Proxy is a special relay mode for proxying requests to custom upstream
Proxy
ImagesEdits
)

2
relay/relaymode/helper.go
View File

@@ -26,6 +26,8 @@ func GetByPath(path string) int {
relayMode = AudioTranslation
} else if strings.HasPrefix(path, "/v1/images/edits") {
relayMode = ImagesEdits
} else if strings.HasPrefix(path, "/v1/oneapi/proxy") {
relayMode = Proxy
}
return relayMode