Skip to content

Commit

Permalink
refactor: use adaptor to do relay & test
Browse files Browse the repository at this point in the history
  • Loading branch information
@songquanpeng
songquanpeng committed Feb 17, 2024
1 parent d548a01 commit 1aa374c
Show file tree
Hide file tree
Showing 63 changed files with 1,441 additions and 1,321 deletions.
111 changes: 60 additions & 51 deletions common/model-ratio.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,12 @@ var DalleImagePromptLengthLimitations = map[string]int{
"dall-e-3": 4000,
}

const (
USD2RMB = 7
USD = 500 // $0.002 = 1 -> $1 = 500
RMB = USD / USD2RMB
)

// ModelRatio
// https://platform.openai.com/docs/models/model-endpoint-compatibility
// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
Expand All @@ -38,57 +44,60 @@ var DalleImagePromptLengthLimitations = map[string]int{
// 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-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-vision-preview": 5, // $0.01 / 1K tokens
"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,
"gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
"gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
"gpt-3.5-turbo-0125": 0.25, // $0.0005 / 1K tokens
"davinci-002": 1, // $0.002 / 1K tokens
"babbage-002": 0.2, // $0.0004 / 1K tokens
"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
"tts-1": 7.5, // $0.015 / 1K characters
"tts-1-1106": 7.5,
"tts-1-hd": 15, // $0.030 / 1K characters
"tts-1-hd-1106": 15,
"davinci": 10,
"curie": 10,
"babbage": 10,
"ada": 10,
"text-embedding-ada-002": 0.05,
"text-embedding-3-small": 0.01,
"text-embedding-3-large": 0.065,
"text-search-ada-doc-001": 10,
"text-moderation-stable": 0.1,
"text-moderation-latest": 0.1,
"dall-e-2": 8, // $0.016 - $0.020 / image
"dall-e-3": 20, // $0.040 - $0.120 / image
"claude-instant-1": 0.815, // $1.63 / 1M tokens
"claude-2": 5.51, // $11.02 / 1M tokens
"claude-2.0": 5.51, // $11.02 / 1M tokens
"claude-2.1": 5.51, // $11.02 / 1M tokens
"ERNIE-Bot": 0.8572, // ￥0.012 / 1k tokens
"ERNIE-Bot-turbo": 0.5715, // ￥0.008 / 1k tokens
"ERNIE-Bot-4": 8.572, // ￥0.12 / 1k tokens
// https://openai.com/pricing
"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-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-vision-preview": 5, // $0.01 / 1K tokens
"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,
"gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
"gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
"gpt-3.5-turbo-0125": 0.25, // $0.0005 / 1K tokens
"davinci-002": 1, // $0.002 / 1K tokens
"babbage-002": 0.2, // $0.0004 / 1K tokens
"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
"tts-1": 7.5, // $0.015 / 1K characters
"tts-1-1106": 7.5,
"tts-1-hd": 15, // $0.030 / 1K characters
"tts-1-hd-1106": 15,
"davinci": 10,
"curie": 10,
"babbage": 10,
"ada": 10,
"text-embedding-ada-002": 0.05,
"text-embedding-3-small": 0.01,
"text-embedding-3-large": 0.065,
"text-search-ada-doc-001": 10,
"text-moderation-stable": 0.1,
"text-moderation-latest": 0.1,
"dall-e-2": 8, // $0.016 - $0.020 / image
"dall-e-3": 20, // $0.040 - $0.120 / image
"claude-instant-1": 0.815, // $1.63 / 1M tokens
"claude-2": 5.51, // $11.02 / 1M tokens
"claude-2.0": 5.51, // $11.02 / 1M tokens
"claude-2.1": 5.51, // $11.02 / 1M tokens
// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7
"ERNIE-Bot": 0.8572, // ￥0.012 / 1k tokens
"ERNIE-Bot-turbo": 0.5715, // ￥0.008 / 1k tokens
"ERNIE-Bot-4": 0.12 * RMB, // ￥0.12 / 1k tokens
"ERNIE-Bot-8k": 0.024 * RMB,
"Embedding-V1": 0.1429, // ￥0.002 / 1k tokens
"PaLM-2": 1,
"gemini-pro": 1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
Expand Down
6 changes: 3 additions & 3 deletions controller/billing.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ import (
"github.com/gin-gonic/gin"
"github.com/songquanpeng/one-api/common/config"
"github.com/songquanpeng/one-api/model"
"github.com/songquanpeng/one-api/relay/channel/openai"
relaymodel "github.com/songquanpeng/one-api/relay/model"
)

func GetSubscription(c *gin.Context) {
Expand All @@ -30,7 +30,7 @@ func GetSubscription(c *gin.Context) {
expiredTime = 0
}
if err != nil {
Error := openai.Error{
Error := relaymodel.Error{
Message: err.Error(),
Type: "upstream_error",
}
Expand Down Expand Up @@ -72,7 +72,7 @@ func GetUsage(c *gin.Context) {
quota, err = model.GetUserUsedQuota(userId)
}
if err != nil {
Error := openai.Error{
Error := relaymodel.Error{
Message: err.Error(),
Type: "one_api_error",
}
Expand Down
138 changes: 62 additions & 76 deletions controller/channel-test.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,103 +9,91 @@ import (
"github.com/songquanpeng/one-api/common/config"
"github.com/songquanpeng/one-api/common/logger"
"github.com/songquanpeng/one-api/model"
"github.com/songquanpeng/one-api/relay/channel/openai"
"github.com/songquanpeng/one-api/relay/constant"
"github.com/songquanpeng/one-api/relay/helper"
relaymodel "github.com/songquanpeng/one-api/relay/model"
"github.com/songquanpeng/one-api/relay/util"
"io"
"net/http"
"net/http/httptest"
"net/url"
"strconv"
"sync"
"time"

"github.com/gin-gonic/gin"
)

func testChannel(channel *model.Channel, request openai.ChatRequest) (err error, openaiErr *openai.Error) {
switch channel.Type {
case common.ChannelTypePaLM:
fallthrough
case common.ChannelTypeGemini:
fallthrough
case common.ChannelTypeAnthropic:
fallthrough
case common.ChannelTypeBaidu:
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:
request.Model = "gpt-35-turbo"
defer func() {
if err != nil {
err = errors.New("请确保已在 Azure 上创建了 gpt-35-turbo 模型，并且 apiVersion 已正确填写！")
}
}()
default:
request.Model = "gpt-3.5-turbo"
}
requestURL := common.ChannelBaseURLs[channel.Type]
if channel.Type == common.ChannelTypeAzure {
requestURL = util.GetFullRequestURL(channel.GetBaseURL(), fmt.Sprintf("/openai/deployments/%s/chat/completions?api-version=2023-03-15-preview", request.Model), channel.Type)
} else {
if baseURL := channel.GetBaseURL(); len(baseURL) > 0 {
requestURL = baseURL
}

requestURL = util.GetFullRequestURL(requestURL, "/v1/chat/completions", channel.Type)
func buildTestRequest() *relaymodel.GeneralOpenAIRequest {
testRequest := &relaymodel.GeneralOpenAIRequest{
MaxTokens: 1,
Stream: false,
Model: "gpt-3.5-turbo",
}
jsonData, err := json.Marshal(request)
if err != nil {
return err, nil
testMessage := relaymodel.Message{
Role: "user",
Content: "hi",
}
req, err := http.NewRequest("POST", requestURL, bytes.NewBuffer(jsonData))
testRequest.Messages = append(testRequest.Messages, testMessage)
return testRequest
}

func testChannel(channel *model.Channel) (err error, openaiErr *relaymodel.Error) {
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
c.Request = &http.Request{
Method: "POST",
URL: &url.URL{Path: "/v1/chat/completions"},
Body: nil,
Header: make(http.Header),
}
c.Request.Header.Set("Authorization", "Bearer "+channel.Key)
c.Request.Header.Set("Content-Type", "application/json")
c.Set("channel", channel.Type)
c.Set("base_url", channel.GetBaseURL())
meta := util.GetRelayMeta(c)
apiType := constant.ChannelType2APIType(channel.Type)
adaptor := helper.GetAdaptor(apiType)
if adaptor == nil {
return fmt.Errorf("invalid api type: %d, adaptor is nil", apiType), nil
}
modelName := adaptor.GetModelList()[0]
request := buildTestRequest()
request.Model = modelName
meta.OriginModelName, meta.ActualModelName = modelName, modelName
convertedRequest, err := adaptor.ConvertRequest(c, constant.RelayModeChatCompletions, request)
if err != nil {
return err, nil
}
if channel.Type == common.ChannelTypeAzure {
req.Header.Set("api-key", channel.Key)
} else {
req.Header.Set("Authorization", "Bearer "+channel.Key)
}
req.Header.Set("Content-Type", "application/json")
resp, err := util.HTTPClient.Do(req)
jsonData, err := json.Marshal(convertedRequest)
if err != nil {
return err, nil
}
defer resp.Body.Close()
var response openai.SlimTextResponse
body, err := io.ReadAll(resp.Body)
requestBody := bytes.NewBuffer(jsonData)
c.Request.Body = io.NopCloser(requestBody)
resp, err := adaptor.DoRequest(c, meta, requestBody)
if err != nil {
return err, nil
}
err = json.Unmarshal(body, &response)
if err != nil {
return fmt.Errorf("Error: %s\nResp body: %s", err, body), nil
if resp.StatusCode != http.StatusOK {
err := util.RelayErrorHandler(resp)
return fmt.Errorf("status code %d: %s", resp.StatusCode, err.Error.Message), &err.Error
}
if response.Usage.CompletionTokens == 0 {
if response.Error.Message == "" {
response.Error.Message = "补全 tokens 非预期返回 0"
}
return fmt.Errorf("type %s, code %v, message %s", response.Error.Type, response.Error.Code, response.Error.Message), &response.Error
usage, respErr := adaptor.DoResponse(c, resp, meta)
if respErr != nil {
return fmt.Errorf("%s", respErr.Error.Message), &respErr.Error
}
return nil, nil
}

func buildTestRequest() *openai.ChatRequest {
testRequest := &openai.ChatRequest{
Model: "", // this will be set later
MaxTokens: 1,
if usage == nil {
return errors.New("usage is nil"), nil
}
testMessage := openai.Message{
Role: "user",
Content: "hi",
result := w.Result()
// print result.Body
respBody, err := io.ReadAll(result.Body)
if err != nil {
return err, nil
}
testRequest.Messages = append(testRequest.Messages, testMessage)
return testRequest
logger.SysLog(fmt.Sprintf("testing channel #%d, response: \n%s", channel.Id, string(respBody)))
return nil, nil
}

func TestChannel(c *gin.Context) {
Expand All @@ -125,9 +113,8 @@ func TestChannel(c *gin.Context) {
})
return
}
testRequest := buildTestRequest()
tik := time.Now()
err, _ = testChannel(channel, *testRequest)
err, _ = testChannel(channel)
tok := time.Now()
milliseconds := tok.Sub(tik).Milliseconds()
go channel.UpdateResponseTime(milliseconds)
Expand Down Expand Up @@ -192,7 +179,6 @@ func testAllChannels(notify bool) error {
if err != nil {
return err
}
testRequest := buildTestRequest()
var disableThreshold = int64(config.ChannelDisableThreshold * 1000)
if disableThreshold == 0 {
disableThreshold = 10000000 // a impossible value
Expand All @@ -201,7 +187,7 @@ func testAllChannels(notify bool) error {
for _, channel := range channels {
isChannelEnabled := channel.Status == common.ChannelStatusEnabled
tik := time.Now()
err, openaiErr := testChannel(channel, *testRequest)
err, openaiErr := testChannel(channel)
tok := time.Now()
milliseconds := tok.Sub(tik).Milliseconds()
if isChannelEnabled && milliseconds > disableThreshold {
Expand Down
Loading

0 comments on commit 1aa374c

Please sign in to comment.