xinyin025/one-api
1
0
Fork 0
mirror of https://github.com/songquanpeng/one-api.git synced 2025-09-17 17:16:38 +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 2024-11-20 02:09:28 +00:00
commit 082cfbe574
42 changed files with 490 additions and 350 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@ -86,6 +86,7 @@ docker image: `ppcelery/one-api:latest`
+ [x] [together.ai](https://www.together.ai/)
+ [x] [novita.ai](https://www.novita.ai/)
+ [x] [硅基流动 SiliconCloud](https://siliconflow.cn/siliconcloud)
+ [x] [xAI](https://x.ai/)
2. 支持配置镜像以及众多[第三方代理服务](https://iamazing.cn/page/openai-api-third-party-services)。
3. 支持通过**负载均衡**的方式访问多个渠道。
4. 支持 **stream 模式**，可以通过流式传输实现打字机效果。
@ -395,6 +396,7 @@ graph LR
26. `METRIC_SUCCESS_RATE_THRESHOLD`：请求成功率阈值，默认为 `0.8`。
27. `INITIAL_ROOT_TOKEN`：如果设置了该值，则在系统首次启动时会自动创建一个值为该环境变量值的 root 用户令牌。
28. `INITIAL_ROOT_ACCESS_TOKEN`：如果设置了该值，则在系统首次启动时会自动创建一个值为该环境变量的 root 用户创建系统管理令牌。
29. `ENFORCE_INCLUDE_USAGE`：是否强制在 stream 模型下返回 usage，默认不开启，可选值为 `true` 和 `false`。
### 命令行参数
1. `--port <port_number>`: 指定服务器监听的端口号，默认为 `3000`。

2
common/config/config.go
View File

@ -175,3 +175,5 @@ var OnlyOneLogFile = env.Bool("ONLY_ONE_LOG_FILE", false)
var RelayProxy = env.String("RELAY_PROXY", "")
var UserContentRequestProxy = env.String("USER_CONTENT_REQUEST_PROXY", "")
var UserContentRequestTimeout = env.Int("USER_CONTENT_REQUEST_TIMEOUT", 30)
var EnforceIncludeUsage = env.Bool("ENFORCE_INCLUDE_USAGE", false)

1
common/ctxkey/key.go
View File

@ -24,4 +24,5 @@ const (
BaseURL = "base_url"
AvailableModels = "available_models"
KeyRequestBody = "key_request_body"
SystemPrompt = "system_prompt"
)

20
common/helper/helper.go
View File

@ -138,3 +138,23 @@ func String2Int(str string) int {
}
return num
}
func Float64PtrMax(p *float64, maxValue float64) *float64 {
if p == nil {
return nil
}
if *p > maxValue {
return &maxValue
}
return p
}
func Float64PtrMin(p *float64, minValue float64) *float64 {
if p == nil {
return nil
}
if *p < minValue {
return &minValue
}
return p
}

3
middleware/distributor.go
View File

@ -85,6 +85,9 @@ func SetupContextForSelectedChannel(c *gin.Context, channel *model.Channel, mode
c.Set(ctxkey.ChannelId, channel.Id)
c.Set(ctxkey.ChannelName, channel.Name)
c.Set(ctxkey.ContentType, c.Request.Header.Get("Content-Type"))
if channel.SystemPrompt != nil && *channel.SystemPrompt != "" {
c.Set(ctxkey.SystemPrompt, *channel.SystemPrompt)
}
c.Set(ctxkey.ModelMapping, channel.GetModelMapping())
c.Set(ctxkey.OriginalModel, modelName) // for retry
c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", channel.Key))

1
model/channel.go
View File

@ -37,6 +37,7 @@ type Channel struct {
ModelMapping *string `json:"model_mapping" gorm:"type:varchar(1024);default:''"`
Priority *int64 `json:"priority" gorm:"bigint;default:0"`
Config string `json:"config"`
SystemPrompt *string `json:"system_prompt" gorm:"type:text"`
}
type ChannelConfig struct {

4
relay/adaptor/ali/main.go
View File

@ -36,9 +36,7 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
enableSearch = true
aliModel = strings.TrimSuffix(aliModel, EnableSearchModelSuffix)
}
if request.TopP >= 1 {
request.TopP = 0.9999
}
request.TopP = helper.Float64PtrMax(request.TopP, 0.9999)
return &ChatRequest{
Model: aliModel,
Input: Input{

4
relay/adaptor/ali/model.go
View File

@ -16,13 +16,13 @@ type Input struct {
}
type Parameters struct {
TopP float64 `json:"top_p,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
TopK int `json:"top_k,omitempty"`
Seed uint64 `json:"seed,omitempty"`
EnableSearch bool `json:"enable_search,omitempty"`
IncrementalOutput bool `json:"incremental_output,omitempty"`
MaxTokens int `json:"max_tokens,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
ResultFormat string `json:"result_format,omitempty"`
Tools []model.Tool `json:"tools,omitempty"`
}

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

@ -3,6 +3,7 @@ package anthropic
var ModelList = []string{
"claude-instant-1.2", "claude-2.0", "claude-2.1",
"claude-3-haiku-20240307",
"claude-3-5-haiku-20241022",
"claude-3-sonnet-20240229",
"claude-3-opus-20240229",
"claude-3-5-sonnet-20240620",

4
relay/adaptor/anthropic/model.go
View File

@ -48,8 +48,8 @@ type Request struct {
MaxTokens int `json:"max_tokens,omitempty"`
StopSequences []string `json:"stop_sequences,omitempty"`
Stream bool `json:"stream,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
TopK int `json:"top_k,omitempty"`
Tools []Tool `json:"tools,omitempty"`
ToolChoice any `json:"tool_choice,omitempty"`

4
relay/adaptor/aws/claude/model.go
View File

@ -11,8 +11,8 @@ type Request struct {
Messages []anthropic.Message `json:"messages"`
System string `json:"system,omitempty"`
MaxTokens int `json:"max_tokens,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
TopK int `json:"top_k,omitempty"`
StopSequences []string `json:"stop_sequences,omitempty"`
Tools []anthropic.Tool `json:"tools,omitempty"`

4
relay/adaptor/aws/llama3/model.go
View File

@ -6,8 +6,8 @@ package aws
type Request struct {
Prompt string `json:"prompt"`
MaxGenLen int `json:"max_gen_len,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
}
// Response is the response from AWS Llama3

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

@ -35,9 +35,9 @@ type Message struct {
type ChatRequest struct {
Messages []Message `json:"messages"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
PenaltyScore float64 `json:"penalty_score,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
PenaltyScore *float64 `json:"penalty_score,omitempty"`
Stream bool `json:"stream,omitempty"`
System string `json:"system,omitempty"`
DisableSearch bool `json:"disable_search,omitempty"`

2
relay/adaptor/cloudflare/model.go
View File

@ -9,5 +9,5 @@ type Request struct {
Prompt string `json:"prompt,omitempty"`
Raw bool `json:"raw,omitempty"`
Stream bool `json:"stream,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
}

2
relay/adaptor/cohere/main.go
View File

@ -43,7 +43,7 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
K: textRequest.TopK,
Stream: textRequest.Stream,
FrequencyPenalty: textRequest.FrequencyPenalty,
PresencePenalty: textRequest.FrequencyPenalty,
PresencePenalty: textRequest.PresencePenalty,
Seed: int(textRequest.Seed),
}
if cohereRequest.Model == "" {

8
relay/adaptor/cohere/model.go
View File

@ -10,15 +10,15 @@ type Request struct {
PromptTruncation string `json:"prompt_truncation,omitempty"` // 默认值为"AUTO"
Connectors []Connector `json:"connectors,omitempty"`
Documents []Document `json:"documents,omitempty"`
Temperature float64 `json:"temperature,omitempty"` // 默认值为0.3
Temperature *float64 `json:"temperature,omitempty"` // 默认值为0.3
MaxTokens int `json:"max_tokens,omitempty"`
MaxInputTokens int `json:"max_input_tokens,omitempty"`
K int `json:"k,omitempty"` // 默认值为0
P float64 `json:"p,omitempty"` // 默认值为0.75
P *float64 `json:"p,omitempty"` // 默认值为0.75
Seed int `json:"seed,omitempty"`
StopSequences []string `json:"stop_sequences,omitempty"`
FrequencyPenalty float64 `json:"frequency_penalty,omitempty"` // 默认值为0.0
PresencePenalty float64 `json:"presence_penalty,omitempty"` // 默认值为0.0
FrequencyPenalty *float64 `json:"frequency_penalty,omitempty"` // 默认值为0.0
PresencePenalty *float64 `json:"presence_penalty,omitempty"` // 默认值为0.0
Tools []Tool `json:"tools,omitempty"`
ToolResults []ToolResult `json:"tool_results,omitempty"`
}

4
relay/adaptor/gemini/model.go
View File

@ -67,8 +67,8 @@ type ChatTools struct {
type ChatGenerationConfig struct {
ResponseMimeType string `json:"responseMimeType,omitempty"`
ResponseSchema any `json:"responseSchema,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"topP,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"topP,omitempty"`
TopK float64 `json:"topK,omitempty"`
MaxOutputTokens int `json:"maxOutputTokens,omitempty"`
CandidateCount int `json:"candidateCount,omitempty"`

8
relay/adaptor/ollama/model.go
View File

@ -2,11 +2,11 @@ package ollama
type Options struct {
Seed int `json:"seed,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopK int `json:"top_k,omitempty"`
TopP float64 `json:"top_p,omitempty"`
FrequencyPenalty float64 `json:"frequency_penalty,omitempty"`
PresencePenalty float64 `json:"presence_penalty,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
FrequencyPenalty *float64 `json:"frequency_penalty,omitempty"`
PresencePenalty *float64 `json:"presence_penalty,omitempty"`
NumPredict int `json:"num_predict,omitempty"`
NumCtx int `json:"num_ctx,omitempty"`
}

6
relay/adaptor/openai/compatible.go
View File

@ -11,9 +11,10 @@ import (
"github.com/songquanpeng/one-api/relay/adaptor/mistral"
"github.com/songquanpeng/one-api/relay/adaptor/moonshot"
"github.com/songquanpeng/one-api/relay/adaptor/novita"
"github.com/songquanpeng/one-api/relay/adaptor/siliconflow"
"github.com/songquanpeng/one-api/relay/adaptor/stepfun"
"github.com/songquanpeng/one-api/relay/adaptor/togetherai"
"github.com/songquanpeng/one-api/relay/adaptor/siliconflow"
"github.com/songquanpeng/one-api/relay/adaptor/xai"
"github.com/songquanpeng/one-api/relay/channeltype"
)
@ -32,6 +33,7 @@ var CompatibleChannels = []int{
channeltype.TogetherAI,
channeltype.Novita,
channeltype.SiliconFlow,
channeltype.XAI,
}
func GetCompatibleChannelMeta(channelType int) (string, []string) {
@ -64,6 +66,8 @@ func GetCompatibleChannelMeta(channelType int) (string, []string) {
return "novita", novita.ModelList
case channeltype.SiliconFlow:
return "siliconflow", siliconflow.ModelList
case channeltype.XAI:
return "xai", xai.ModelList
default:
return "openai", ModelList
}

4
relay/adaptor/palm/model.go
View File

@ -20,9 +20,9 @@ type Prompt struct {
type ChatRequest struct {
Prompt Prompt `json:"prompt"`
Temperature float64 `json:"temperature,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
CandidateCount int `json:"candidateCount,omitempty"`
TopP float64 `json:"topP,omitempty"`
TopP *float64 `json:"topP,omitempty"`
TopK int `json:"topK,omitempty"`
}

4
relay/adaptor/tencent/main.go
View File

@ -39,8 +39,8 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
Model: &request.Model,
Stream: &request.Stream,
Messages: messages,
TopP: &request.TopP,
Temperature: &request.Temperature,
TopP: request.TopP,
Temperature: request.Temperature,
}
}

4
relay/adaptor/vertexai/claude/model.go
View File

@ -11,8 +11,8 @@ type Request struct {
MaxTokens int `json:"max_tokens,omitempty"`
StopSequences []string `json:"stop_sequences,omitempty"`
Stream bool `json:"stream,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
TopK int `json:"top_k,omitempty"`
Tools []anthropic.Tool `json:"tools,omitempty"`
ToolChoice any `json:"tool_choice,omitempty"`

5
relay/adaptor/xai/constants.go Normal file
View File

@ -0,0 +1,5 @@
package xai
var ModelList = []string{
"grok-beta",
}

2
relay/adaptor/xunfei/main.go
View File

@ -283,7 +283,7 @@ func parseAPIVersionByModelName(modelName string) string {
func apiVersion2domain(apiVersion string) string {
switch apiVersion {
case "v1.1":
return "general"
return "lite"
case "v2.1":
return "generalv2"
case "v3.1":

2
relay/adaptor/xunfei/model.go
View File

@ -20,7 +20,7 @@ type ChatRequest struct {
Parameter struct {
Chat struct {
Domain string `json:"domain,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopK int `json:"top_k,omitempty"`
MaxTokens int `json:"max_tokens,omitempty"`
Auditing bool `json:"auditing,omitempty"`

14
relay/adaptor/zhipu/adaptor.go
View File

@ -4,11 +4,11 @@ import (
"errors"
"fmt"
"io"
"math"
"net/http"
"strings"
"github.com/gin-gonic/gin"
"github.com/songquanpeng/one-api/common/helper"
"github.com/songquanpeng/one-api/relay/adaptor"
"github.com/songquanpeng/one-api/relay/adaptor/openai"
"github.com/songquanpeng/one-api/relay/meta"
@ -66,13 +66,13 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
baiduEmbeddingRequest, err := ConvertEmbeddingRequest(*request)
return baiduEmbeddingRequest, err
default:
// TopP (0.0, 1.0)
request.TopP = math.Min(0.99, request.TopP)
request.TopP = math.Max(0.01, request.TopP)
// TopP [0.0, 1.0]
request.TopP = helper.Float64PtrMax(request.TopP, 1)
request.TopP = helper.Float64PtrMin(request.TopP, 0)
// Temperature (0.0, 1.0)
request.Temperature = math.Min(0.99, request.Temperature)
request.Temperature = math.Max(0.01, request.Temperature)
// Temperature [0.0, 1.0]
request.Temperature = helper.Float64PtrMax(request.Temperature, 1)
request.Temperature = helper.Float64PtrMin(request.Temperature, 0)
a.SetVersionByModeName(request.Model)
if a.APIVersion == "v4" {
return request, nil

4
relay/adaptor/zhipu/model.go
View File

@ -13,8 +13,8 @@ type Message struct {
type Request struct {
Prompt []Message `json:"prompt"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
RequestId string `json:"request_id,omitempty"`
Incremental bool `json:"incremental,omitempty"`
}

5
relay/billing/ratio/model.go
View File

@ -79,6 +79,7 @@ var ModelRatio = map[string]float64{
"claude-2.0": 8.0 / 1000 * USD,
"claude-2.1": 8.0 / 1000 * USD,
"claude-3-haiku-20240307": 0.25 / 1000 * USD,
"claude-3-5-haiku-20241022": 1.0 / 1000 * USD,
"claude-3-sonnet-20240229": 3.0 / 1000 * USD,
"claude-3-5-sonnet-20240620": 3.0 / 1000 * USD,
"claude-3-5-sonnet-20241022": 3.0 / 1000 * USD,
@ -208,6 +209,8 @@ var ModelRatio = map[string]float64{
"deepl-zh": 25.0 / 1000 * USD,
"deepl-en": 25.0 / 1000 * USD,
"deepl-ja": 25.0 / 1000 * USD,
// https://console.x.ai/
"grok-beta": 5.0 / 1000 * USD,
}
var CompletionRatio = map[string]float64{
@ -372,6 +375,8 @@ func GetCompletionRatio(name string, channelType int) float64 {
return 3
case "command-r-plus":
return 5
case "grok-beta":
return 3
}
return 1
}

1
relay/channeltype/define.go
View File

@ -46,5 +46,6 @@ const (
VertextAI
Proxy
SiliconFlow
XAI
Dummy
)

1
relay/channeltype/url.go
View File

@ -46,6 +46,7 @@ var ChannelBaseURLs = []string{
"", // 42
"", // 43
"https://api.siliconflow.cn", // 44
"https://api.x.ai", // 45
}
func init() {

1
relay/constant/role/define.go
View File

@ -1,5 +1,6 @@
package role
const (
System = "system"
Assistant = "assistant"
)

41
relay/controller/helper.go
View File

@ -3,8 +3,12 @@ package controller
import (
"context"
"fmt"
"github.com/Laisky/errors/v2"
"math"
"net/http"
"strings"
"github.com/gin-gonic/gin"
"github.com/pkg/errors"
"github.com/songquanpeng/one-api/common"
"github.com/songquanpeng/one-api/common/config"
"github.com/songquanpeng/one-api/common/logger"
@ -12,13 +16,11 @@ import (
"github.com/songquanpeng/one-api/relay/adaptor/openai"
billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
"github.com/songquanpeng/one-api/relay/channeltype"
"github.com/songquanpeng/one-api/relay/constant/role"
"github.com/songquanpeng/one-api/relay/controller/validator"
"github.com/songquanpeng/one-api/relay/meta"
relaymodel "github.com/songquanpeng/one-api/relay/model"
"github.com/songquanpeng/one-api/relay/relaymode"
"math"
"net/http"
"strings"
)
func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.GeneralOpenAIRequest, error) {
@ -89,7 +91,7 @@ func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIR
return preConsumedQuota, nil
}
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64) (quota int64) {
func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64, systemPromptReset bool) {
if usage == nil {
logger.Error(ctx, "usage is nil, which is unexpected")
return
@ -98,7 +100,7 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
completionRatio := billingratio.GetCompletionRatio(textRequest.Model, meta.ChannelType)
promptTokens := usage.PromptTokens
completionTokens := usage.CompletionTokens
quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
quota := int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
if ratio != 0 && quota <= 0 {
quota = 1
}
@ -117,8 +119,11 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
if err != nil {
logger.Error(ctx, "error update user quota cache: "+err.Error())
}
logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f", modelRatio, groupRatio, completionRatio)
var extraLog string
if systemPromptReset {
extraLog = " （注意系统提示词已被重置）"
}
logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f，补全倍率 %.2f%s", modelRatio, groupRatio, completionRatio, extraLog)
model.RecordConsumeLog(ctx, meta.UserId, meta.ChannelId, promptTokens, completionTokens, textRequest.Model, meta.TokenName, quota, logContent)
model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
model.UpdateChannelUsedQuota(meta.ChannelId, quota)
@ -156,3 +161,23 @@ func isErrorHappened(meta *meta.Meta, resp *http.Response) bool {
}
return false
}
func setSystemPrompt(ctx context.Context, request *relaymodel.GeneralOpenAIRequest, prompt string) (reset bool) {
if prompt == "" {
return false
}
if len(request.Messages) == 0 {
return false
}
if request.Messages[0].Role == role.System {
request.Messages[0].Content = prompt
logger.Infof(ctx, "rewrite system prompt")
return true
}
request.Messages = append([]relaymodel.Message{{
Role: role.System,
Content: prompt,
}}, request.Messages...)
logger.Infof(ctx, "add system prompt")
return true
}

19
relay/controller/text.go
View File

@ -8,9 +8,9 @@ import (
"github.com/Laisky/errors/v2"
"github.com/gin-gonic/gin"
"github.com/songquanpeng/one-api/common/config"
"github.com/songquanpeng/one-api/common/ctxkey"
"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"
@ -37,6 +37,8 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
meta.OriginModelName = textRequest.Model
textRequest.Model, _ = getMappedModelName(textRequest.Model, meta.ModelMapping)
meta.ActualModelName = textRequest.Model
// set system prompt if not empty
systemPromptReset := setSystemPrompt(ctx, textRequest, meta.SystemPrompt)
// get model ratio & group ratio
modelRatio := billingratio.GetModelRatio(textRequest.Model, meta.ChannelType)
// groupRatio := billingratio.GetGroupRatio(meta.Group)
@ -88,23 +90,12 @@ func RelayTextHelper(c *gin.Context) *relaymodel.ErrorWithStatusCode {
}
// post-consume quota
go func() {
quota := postConsumeQuota(c, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio)
docu := model.NewUserRequestCost(
c.GetInt(ctxkey.Id),
c.GetString(ctxkey.RequestId),
quota,
)
if err = docu.Insert(); err != nil {
logger.Errorf(c, "insert user request cost failed: %+v", err)
}
}()
go postConsumeQuota(ctx, usage, meta, textRequest, ratio, preConsumedQuota, modelRatio, groupRatio, systemPromptReset)
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 {
if !config.EnforceIncludeUsage && meta.APIType == apitype.OpenAI && meta.OriginModelName == meta.ActualModelName && meta.ChannelType != channeltype.Baichuan {
// no need to convert request for openai
return c.Request.Body, nil
}

2
relay/meta/relay_meta.go
View File

@ -32,6 +32,7 @@ type Meta struct {
RequestURLPath string
PromptTokens int // only for DoResponse
ChannelRatio float64
SystemPrompt string
}
func GetByContext(c *gin.Context) *Meta {
@ -49,6 +50,7 @@ func GetByContext(c *gin.Context) *Meta {
APIKey: strings.TrimPrefix(c.Request.Header.Get("Authorization"), "Bearer "),
RequestURLPath: c.Request.URL.String(),
ChannelRatio: c.GetFloat64(ctxkey.ChannelRatio), // add by Laisky
SystemPrompt: c.GetString(ctxkey.SystemPrompt),
}
cfg, ok := c.Get(ctxkey.Config)
if ok {

33
relay/model/general.go
View File

@ -22,33 +22,48 @@ type StreamOptions struct {
}
type GeneralOpenAIRequest struct {
// https://platform.openai.com/docs/api-reference/chat/create
Messages []Message `json:"messages,omitempty"`
Model string `json:"model,omitempty"`
Modalities []string `json:"modalities,omitempty"`
Audio *Audio `json:"audio,omitempty"`
FrequencyPenalty float64 `json:"frequency_penalty,omitempty"`
Store *bool `json:"store,omitempty"`
Metadata any `json:"metadata,omitempty"`
FrequencyPenalty *float64 `json:"frequency_penalty,omitempty"`
LogitBias any `json:"logit_bias,omitempty"`
Logprobs *bool `json:"logprobs,omitempty"`
TopLogprobs *int `json:"top_logprobs,omitempty"`
MaxTokens int `json:"max_tokens,omitempty"`
MaxCompletionTokens *int `json:"max_completion_tokens,omitempty"`
N int `json:"n,omitempty"`
PresencePenalty float64 `json:"presence_penalty,omitempty"`
Modalities []string `json:"modalities,omitempty"`
Prediction any `json:"prediction,omitempty"`
Audio *Audio `json:"audio,omitempty"`
PresencePenalty *float64 `json:"presence_penalty,omitempty"`
ResponseFormat *ResponseFormat `json:"response_format,omitempty"`
Seed float64 `json:"seed,omitempty"`
ServiceTier *string `json:"service_tier,omitempty"`
Stop any `json:"stop,omitempty"`
Stream bool `json:"stream,omitempty"`
StreamOptions *StreamOptions `json:"stream_options,omitempty"`
Temperature float64 `json:"temperature,omitempty"`
TopP float64 `json:"top_p,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
TopP *float64 `json:"top_p,omitempty"`
TopK int `json:"top_k,omitempty"`
Tools []Tool `json:"tools,omitempty"`
ToolChoice any `json:"tool_choice,omitempty"`
ParallelTooCalls *bool `json:"parallel_tool_calls,omitempty"`
User string `json:"user,omitempty"`
FunctionCall any `json:"function_call,omitempty"`
Functions any `json:"functions,omitempty"`
User string `json:"user,omitempty"`
Prompt any `json:"prompt,omitempty"`
// https://platform.openai.com/docs/api-reference/embeddings/create
Input any `json:"input,omitempty"`
EncodingFormat string `json:"encoding_format,omitempty"`
Dimensions int `json:"dimensions,omitempty"`
Instruction string `json:"instruction,omitempty"`
// https://platform.openai.com/docs/api-reference/images/create
Prompt any `json:"prompt,omitempty"`
Quality *string `json:"quality,omitempty"`
Size string `json:"size,omitempty"`
Style *string `json:"style,omitempty"`
// Others
Instruction string `json:"instruction,omitempty"`
NumCtx int `json:"num_ctx,omitempty"`
}

1
web/air/src/constants/channel.constants.js
View File

@ -30,6 +30,7 @@ export const CHANNEL_OPTIONS = [
{ key: 42, text: 'VertexAI', value: 42, color: 'blue' },
{ key: 43, text: 'Proxy', value: 43, color: 'blue' },
{ key: 44, text: 'SiliconFlow', value: 44, color: 'blue' },
{ key: 45, text: 'xAI', value: 45, color: 'blue' },
{ key: 8, text: '自定义渠道', value: 8, color: 'pink' },
{ key: 22, text: '知识库：FastGPT', value: 22, color: 'blue' },
{ key: 21, text: '知识库：AI Proxy', value: 21, color: 'purple' },

16
web/air/src/pages/Channel/EditChannel.js
View File

@ -43,6 +43,7 @@ const EditChannel = (props) => {
base_url: '',
other: '',
model_mapping: '',
system_prompt: '',
models: [],
auto_ban: 1,
groups: ['default']
@ -63,7 +64,7 @@ const EditChannel = (props) => {
let localModels = [];
switch (value) {
case 14:
localModels = ["claude-instant-1.2", "claude-2", "claude-2.0", "claude-2.1", "claude-3-opus-20240229", "claude-3-sonnet-20240229", "claude-3-haiku-20240307", "claude-3-5-sonnet-20240620", "claude-3-5-sonnet-20241022"];
localModels = ["claude-instant-1.2", "claude-2", "claude-2.0", "claude-2.1", "claude-3-opus-20240229", "claude-3-sonnet-20240229", "claude-3-haiku-20240307", "claude-3-5-haiku-20241022", "claude-3-5-sonnet-20240620", "claude-3-5-sonnet-20241022"];
break;
case 11:
localModels = ['PaLM-2'];
@ -496,6 +497,19 @@ const EditChannel = (props) => {
value={inputs.model_mapping}
autoComplete='new-password'
/>
<div style={{ marginTop: 10 }}>
<Typography.Text strong>系统提示词：</Typography.Text>
</div>
<TextArea
placeholder={`此项可选，用于强制设置给定的系统提示词，请配合自定义模型 & 模型重定向使用，首先创建一个唯一的自定义模型名称并在上面填入，之后将该自定义模型重定向映射到该渠道一个原生支持的模型`}
name='system_prompt'
onChange={value => {
handleInputChange('system_prompt', value)
}}
autosize
value={inputs.system_prompt}
autoComplete='new-password'
/>
<Typography.Text style={{
color: 'rgba(var(--semi-blue-5), 1)',
userSelect: 'none',

6
web/berry/src/constants/ChannelConstants.js
View File

@ -179,6 +179,12 @@ export const CHANNEL_OPTIONS = {
value: 44,
color: 'primary'
},
45: {
key: 45,
text: 'xAI',
value: 45,
color: 'primary'
},
41: {
key: 41,
text: 'Novita',

22
web/berry/src/views/Channel/component/EditModal.js
View File

@ -595,6 +595,28 @@ const EditModal = ({ open, channelId, onCancel, onOk }) => {
<FormHelperText id="helper-tex-channel-model_mapping-label"> {inputPrompt.model_mapping} </FormHelperText>
)}
</FormControl>
<FormControl fullWidth error={Boolean(touched.system_prompt && errors.system_prompt)} sx={{ ...theme.typography.otherInput }}>
{/* <InputLabel htmlFor="channel-model_mapping-label">{inputLabel.model_mapping}</InputLabel> */}
<TextField
multiline
id="channel-system_prompt-label"
label={inputLabel.system_prompt}
value={values.system_prompt}
name="system_prompt"
onBlur={handleBlur}
onChange={handleChange}
aria-describedby="helper-text-channel-system_prompt-label"
minRows={5}
placeholder={inputPrompt.system_prompt}
/>
{touched.system_prompt && errors.system_prompt ? (
<FormHelperText error id="helper-tex-channel-system_prompt-label">
{errors.system_prompt}
</FormHelperText>
) : (
<FormHelperText id="helper-tex-channel-system_prompt-label"> {inputPrompt.system_prompt} </FormHelperText>
)}
</FormControl>
<DialogActions>
<Button onClick={onCancel}>取消</Button>
<Button disableElevation disabled={isSubmitting} type="submit" variant="contained" color="primary">

5
web/berry/src/views/Channel/type/Config.js
View File

@ -18,6 +18,7 @@ const defaultConfig = {
other: '其他参数',
models: '模型',
model_mapping: '模型映射关系',
system_prompt: '系统提示词',
groups: '用户组',
config: null
},
@ -30,6 +31,7 @@ const defaultConfig = {
models: '请选择该渠道所支持的模型',
model_mapping:
'请输入要修改的模型映射关系，格式为：api请求模型ID:实际转发给渠道的模型ID，使用JSON数组表示，例如：{"gpt-3.5": "gpt-35"}',
system_prompt:"此项可选，用于强制设置给定的系统提示词，请配合自定义模型 & 模型重定向使用，首先创建一个唯一的自定义模型名称并在上面填入，之后将该自定义模型重定向映射到该渠道一个原生支持的模型此项可选，用于强制设置给定的系统提示词，请配合自定义模型 & 模型重定向使用，首先创建一个唯一的自定义模型名称并在上面填入，之后将该自定义模型重定向映射到该渠道一个原生支持的模型",
groups: '请选择该渠道所支持的用户组',
config: null
},
@ -223,6 +225,9 @@ const typeConfig = {
},
modelGroup: 'anthropic'
},
45: {
modelGroup: 'xai'
},
};
export { defaultConfig, typeConfig };

1
web/default/src/constants/channel.constants.js
View File

@ -30,6 +30,7 @@ export const CHANNEL_OPTIONS = [
{ key: 42, text: 'VertexAI', value: 42, color: 'blue' },
{ key: 43, text: 'Proxy', value: 43, color: 'blue' },
{ key: 44, text: 'SiliconFlow', value: 44, color: 'blue' },
{ key: 45, text: 'xAI', value: 45, color: 'blue' },
{ key: 8, text: '自定义渠道', value: 8, color: 'pink' },
{ key: 22, text: '知识库：FastGPT', value: 22, color: 'blue' },
{ key: 21, text: '知识库：AI Proxy', value: 21, color: 'purple' },

15
web/default/src/pages/Channel/EditChannel.js
View File

@ -43,6 +43,7 @@ const EditChannel = () => {
base_url: '',
other: '',
model_mapping: '',
system_prompt: '',
models: [],
groups: ['default']
};
@ -425,7 +426,7 @@ const EditChannel = () => {
)
}
{
inputs.type !== 43 && (
inputs.type !== 43 && (<>
<Form.Field>
<Form.TextArea
label='模型重定向'
@ -437,6 +438,18 @@ const EditChannel = () => {
autoComplete='new-password'
/>
</Form.Field>
<Form.Field>
<Form.TextArea
label='系统提示词'
placeholder={`此项可选，用于强制设置给定的系统提示词，请配合自定义模型 & 模型重定向使用，首先创建一个唯一的自定义模型名称并在上面填入，之后将该自定义模型重定向映射到该渠道一个原生支持的模型`}
name='system_prompt'
onChange={handleInputChange}
value={inputs.system_prompt}
style={{ minHeight: 150, fontFamily: 'JetBrains Mono, Consolas' }}
autoComplete='new-password'
/>
</Form.Field>
</>
)
}
{