Compare commits

...

51 Commits

Author SHA1 Message Date
Calcium-Ion
40e640511b Merge pull request #1139 from RedwindA/gemini-fix
feat: 增加对Gemini MimeType类型的验证
2025-06-02 22:33:01 +08:00
Calcium-Ion
5930bb88bf Merge pull request #1140 from RedwindA/gemini-tool-fix
fix: 完善Gemini渠道对tools中additionalProperties的清理
2025-06-02 22:32:43 +08:00
Calcium-Ion
8948e99eeb Merge pull request #1141 from xqx121/patch-1
Fix: The edit interface is not billed (usage-based pricing).
2025-06-02 22:32:18 +08:00
xqx121
37caafc722 Fix: The edit interface is not billed (usage-based pricing). 2025-06-02 22:11:11 +08:00
RedwindA
148c974912 feat: 增加对GeminiMIME类型的验证 2025-06-02 19:00:55 +08:00
RedwindA
f1ee9a301d refactor: enhance cleanFunctionParameters for improved handling of JSON schema, including support for $defs and conditional keywords 2025-06-01 02:08:13 +08:00
CaIon
611d77e1a9 feat: add ToMap method and enhance OpenAI request handling 2025-06-01 01:10:10 +08:00
Calcium-Ion
b05bb899f1 Merge pull request #1134 from QuantumNous/fix_ping_keepalive
fix: 流式请求ping
2025-05-31 22:16:16 +08:00
creamlike1024
c51a30b862 fix: 流式请求ping 2025-05-31 22:13:17 +08:00
Calcium-Ion
9c4d3a6359 Merge pull request #1122 from akkuman/feat/stream-tts
feat: streaming response for tts
2025-05-31 18:44:48 +08:00
Calcium-Ion
6936a795a6 Merge pull request #1123 from RedwindA/patch-3
Add `ERROR_LOG_ENABLED` description in README
2025-05-31 18:44:24 +08:00
Calcium-Ion
74defce481 Merge pull request #1130 from xqx121/main
Fix: Gemini2.5pro ThinkingConfig
2025-05-31 18:43:57 +08:00
xqx121
1c4d7fd84b Fix: Gemini2.5pro ThinkingConfig 2025-05-31 17:50:00 +08:00
RedwindA
f907c25b21 Add ERROR_LOG_ENABLED description 2025-05-29 12:35:13 +08:00
RedwindA
1b64db5521 Add ERROR_LOG_ENABLED description 2025-05-29 12:33:27 +08:00
Akkuman
d608a6f123 feat: streaming response for tts 2025-05-29 10:56:01 +08:00
creamlike1024
361b0abec9 fix: pingerCtx 泄漏 2025-05-28 21:34:45 +08:00
CaIon
e01b517843 fix: Change ParallelTooCalls from bool to *bool in GeneralOpenAIRequest for optional handling 2025-05-28 21:12:55 +08:00
CaIon
f613a79f3e feat: Enhance image request validation in relay-image.go: set default model and size, improve error handling for size format, and ensure prompt and N parameters are validated correctly. 2025-05-28 20:18:37 +08:00
IcedTangerine
87540b4f7c Merge pull request #1110 from wangr0031/fix_parallel_tool_calls
feat: chat/completion路由透传parallel_tool_calls参数
2025-05-28 14:25:43 +08:00
IcedTangerine
e3d7b31a49 Update openai_request.go 2025-05-28 14:25:24 +08:00
IcedTangerine
bf016543c3 Merge pull request #1113 from tbphp/tbphp_vertex_gemini_global_region
fix: Vertex channel global region format
2025-05-28 14:16:47 +08:00
IcedTangerine
eb94aa13e6 Merge pull request #1111 from feitianbubu/fxm-ali-fetch-models-url
fix: ali FetchUpstreamModels url
2025-05-28 14:11:17 +08:00
tbphp
6e72dcd0ba fix: Vertex channel global region format 2025-05-27 21:50:53 +08:00
skynono
96ab4177ca fix: ali FetchUpstreamModels url 2025-05-27 11:22:40 +08:00
wang.rong
76824a0337 chat/completion透传parallel_tool_calls参数 2025-05-27 09:32:20 +08:00
IcedTangerine
3cd29a4963 Merge pull request #1109 from feitianbubu/fix-qwen-thinking
fix: ali parameter.enable_thinking must be set to false for non-strea…
2025-05-26 19:32:34 +08:00
creamlike1024
41120b4d75 Merge branch 'main' of github.com:QuantumNous/new-api 2025-05-26 18:56:14 +08:00
creamlike1024
30d5a11f46 fix: search-preview model web search billing 2025-05-26 18:53:41 +08:00
skynono
368fd75c86 fix: ali parameter.enable_thinking must be set to false for non-streaming calls 2025-05-26 17:41:02 +08:00
IcedTangerine
ee07762611 Merge pull request #1075 from feitianbubu/fix-default-model-not-exist
fix: if default model is not exist, set the first one as default
2025-05-26 17:21:14 +08:00
IcedTangerine
a215538b4d Merge pull request #1081 from feitianbubu/fixTypoOidcEnabledField
fix: typo in oidc_enabled field (previously oidc)
2025-05-26 17:20:35 +08:00
IcedTangerine
873e3f3dc8 Merge pull request #1099 from feitianbubu/fixTagModeStatusSave
fix: keep BatchDelete and TagMode enabled status
2025-05-26 17:17:34 +08:00
Calcium-Ion
b564cac048 Merge pull request #1100 from daggeryu/patch-4
fix aws claude-sonnet-4-20250514
2025-05-24 15:27:30 +08:00
CaIon
fbdad581b5 fix: improve input validation and error handling in ModelSetting and SettingGeminiModel components 2025-05-24 15:26:55 +08:00
daggeryu
0595636ceb fix aws claude-sonnet-4-20250514 2025-05-24 01:21:14 +08:00
CaIon
d95c2436d7 feat: add support for new regions in Claude Sonnet 4 and Claude Opus 4 models in AWS constants 2025-05-23 21:11:00 +08:00
skynono
2cc2d4f652 fix: keep BatchDelete and TagMode enabled status 2025-05-23 20:17:48 +08:00
CaIon
1644b7b15d feat: add new model entries for Claude Sonnet 4 and Claude Opus 4 across multiple components, including constants and cache settings 2025-05-23 15:20:16 +08:00
CaIon
66a8612d12 feat: add new model ratios for Claude Sonnet 4 and Claude Opus 4; update ratio retrieval logic for improved handling of model names 2025-05-23 02:02:21 +08:00
CaIon
f796c3b216 fix: update Init method to correctly set RequestMode based on upstream model name prefixes 2025-05-23 01:34:53 +08:00
CaIon
c53a48cde5 feat: add panic recovery and retry mechanism for InitChannelCache; improve batch deletion of abilities in FixAbility 2025-05-23 01:26:52 +08:00
CaIon
9a59da16a5 feat: implement search functionality in ChannelsTable for improved channel filtering 2025-05-22 16:54:55 +08:00
CaIon
e18001299b feat: enhance Gemini response handling by adding reasoning content and updating JSON decoding method 2025-05-22 16:11:50 +08:00
CaIon
66bdfe180c feat: add Thought field to GeminiPart and update response handling in streamResponseGeminiChat2OpenAI 2025-05-22 15:52:23 +08:00
skynono
e1190f98e9 fix: typo in oidc_enabled field (previously oidc) 2025-05-21 09:33:57 +08:00
CaIon
1f9fc09989 feat: add OutputFormat field to ImageRequest for enhanced image processing options 2025-05-20 19:40:29 +08:00
CaIon
498d73f67c refactor: update JSON field names in GeminiChatRequest for consistency 2025-05-19 20:26:30 +08:00
skynono
9c12e02cb5 fix: if default model is not exist, set the first one as default 2025-05-19 14:56:39 +08:00
IcedTangerine
0ca17d3e6d Merge pull request #1071 from feitianbubu/fixMjImageProxy
fix: proxy settings not applied when request MJ image url
2025-05-18 14:56:47 +08:00
skynono
9927e5d191 fix: proxy settings not applied when request MJ image url 2025-05-16 18:07:56 +08:00
29 changed files with 576 additions and 306 deletions

View File

@@ -110,6 +110,7 @@ For detailed configuration instructions, please refer to [Installation Guide-Env
- `AZURE_DEFAULT_API_VERSION`: Azure channel default API version, default is `2025-04-01-preview`
- `NOTIFICATION_LIMIT_DURATION_MINUTE`: Notification limit duration, default is `10` minutes
- `NOTIFY_LIMIT_COUNT`: Maximum number of user notifications within the specified duration, default is `2`
- `ERROR_LOG_ENABLED=true`: Whether to record and display error logs, default is `false`
## Deployment

View File

@@ -110,6 +110,7 @@ New API提供了丰富的功能详细特性请参考[特性说明](https://do
- `AZURE_DEFAULT_API_VERSION`Azure渠道默认API版本默认 `2025-04-01-preview`
- `NOTIFICATION_LIMIT_DURATION_MINUTE`:通知限制持续时间,默认 `10`分钟
- `NOTIFY_LIMIT_COUNT`:用户通知在指定持续时间内的最大数量,默认 `2`
- `ERROR_LOG_ENABLED=true`: 是否记录并显示错误日志,默认`false`
## 部署

View File

@@ -119,8 +119,11 @@ func FetchUpstreamModels(c *gin.Context) {
baseURL = channel.GetBaseURL()
}
url := fmt.Sprintf("%s/v1/models", baseURL)
if channel.Type == common.ChannelTypeGemini {
switch channel.Type {
case common.ChannelTypeGemini:
url = fmt.Sprintf("%s/v1beta/openai/models", baseURL)
case common.ChannelTypeAli:
url = fmt.Sprintf("%s/compatible-mode/v1/models", baseURL)
}
body, err := GetResponseBody("GET", url, channel, GetAuthHeader(channel.Key))
if err != nil {

View File

@@ -14,6 +14,7 @@ type ImageRequest struct {
ExtraFields json.RawMessage `json:"extra_fields,omitempty"`
Background string `json:"background,omitempty"`
Moderation string `json:"moderation,omitempty"`
OutputFormat string `json:"output_format,omitempty"`
}
type ImageResponse struct {

View File

@@ -2,6 +2,7 @@ package dto
import (
"encoding/json"
"one-api/common"
"strings"
)
@@ -43,6 +44,7 @@ type GeneralOpenAIRequest struct {
ResponseFormat *ResponseFormat `json:"response_format,omitempty"`
EncodingFormat any `json:"encoding_format,omitempty"`
Seed float64 `json:"seed,omitempty"`
ParallelTooCalls *bool `json:"parallel_tool_calls,omitempty"`
Tools []ToolCallRequest `json:"tools,omitempty"`
ToolChoice any `json:"tool_choice,omitempty"`
User string `json:"user,omitempty"`
@@ -53,6 +55,14 @@ type GeneralOpenAIRequest struct {
Audio any `json:"audio,omitempty"`
EnableThinking any `json:"enable_thinking,omitempty"` // ali
ExtraBody any `json:"extra_body,omitempty"`
WebSearchOptions *WebSearchOptions `json:"web_search_options,omitempty"`
}
func (r *GeneralOpenAIRequest) ToMap() map[string]any {
result := make(map[string]any)
data, _ := common.EncodeJson(r)
_ = common.DecodeJson(data, &result)
return result
}
type ToolCallRequest struct {
@@ -72,11 +82,11 @@ type StreamOptions struct {
IncludeUsage bool `json:"include_usage,omitempty"`
}
func (r GeneralOpenAIRequest) GetMaxTokens() int {
func (r *GeneralOpenAIRequest) GetMaxTokens() int {
return int(r.MaxTokens)
}
func (r GeneralOpenAIRequest) ParseInput() []string {
func (r *GeneralOpenAIRequest) ParseInput() []string {
if r.Input == nil {
return nil
}
@@ -371,6 +381,11 @@ func (m *Message) ParseContent() []MediaContent {
return contentList
}
type WebSearchOptions struct {
SearchContextSize string `json:"search_context_size,omitempty"`
UserLocation json.RawMessage `json:"user_location,omitempty"`
}
type OpenAIResponsesRequest struct {
Model string `json:"model"`
Input json.RawMessage `json:"input,omitempty"`

19
main.go
View File

@@ -89,9 +89,22 @@ func main() {
if common.MemoryCacheEnabled {
common.SysLog("memory cache enabled")
common.SysError(fmt.Sprintf("sync frequency: %d seconds", common.SyncFrequency))
model.InitChannelCache()
}
if common.MemoryCacheEnabled {
// Add panic recovery and retry for InitChannelCache
func() {
defer func() {
if r := recover(); r != nil {
common.SysError(fmt.Sprintf("InitChannelCache panic: %v, retrying once", r))
// Retry once
_, fixErr := model.FixAbility()
if fixErr != nil {
common.SysError(fmt.Sprintf("InitChannelCache failed: %s", fixErr.Error()))
}
}
}()
model.InitChannelCache()
}()
go model.SyncOptions(common.SyncFrequency)
go model.SyncChannelCache(common.SyncFrequency)
}

View File

@@ -50,7 +50,7 @@ func getPriority(group string, model string, retry int) (int, error) {
err := DB.Model(&Ability{}).
Select("DISTINCT(priority)").
Where(groupCol+" = ? and model = ? and enabled = "+trueVal, group, model).
Order("priority DESC"). // 按优先级降序排序
Order("priority DESC"). // 按优先级降序排序
Pluck("priority", &priorities).Error // Pluck用于将查询的结果直接扫描到一个切片中
if err != nil {
@@ -261,12 +261,28 @@ func FixAbility() (int, error) {
common.SysError(fmt.Sprintf("Get channel ids from channel table failed: %s", err.Error()))
return 0, err
}
// Delete abilities of channels that are not in channel table
err = DB.Where("channel_id NOT IN (?)", channelIds).Delete(&Ability{}).Error
if err != nil {
common.SysError(fmt.Sprintf("Delete abilities of channels that are not in channel table failed: %s", err.Error()))
return 0, err
// Delete abilities of channels that are not in channel table - in batches to avoid too many placeholders
if len(channelIds) > 0 {
// Process deletion in chunks to avoid "too many placeholders" error
for _, chunk := range lo.Chunk(channelIds, 100) {
err = DB.Where("channel_id NOT IN (?)", chunk).Delete(&Ability{}).Error
if err != nil {
common.SysError(fmt.Sprintf("Delete abilities of channels (batch) that are not in channel table failed: %s", err.Error()))
return 0, err
}
}
} else {
// If no channels exist, delete all abilities
err = DB.Delete(&Ability{}).Error
if err != nil {
common.SysError(fmt.Sprintf("Delete all abilities failed: %s", err.Error()))
return 0, err
}
common.SysLog("Delete all abilities successfully")
return 0, nil
}
common.SysLog(fmt.Sprintf("Delete abilities of channels that are not in channel table successfully, ids: %v", channelIds))
count += len(channelIds)
@@ -275,17 +291,26 @@ func FixAbility() (int, error) {
err = DB.Table("abilities").Distinct("channel_id").Pluck("channel_id", &abilityChannelIds).Error
if err != nil {
common.SysError(fmt.Sprintf("Get channel ids from abilities table failed: %s", err.Error()))
return 0, err
return count, err
}
var channels []Channel
if len(abilityChannelIds) == 0 {
err = DB.Find(&channels).Error
} else {
err = DB.Where("id NOT IN (?)", abilityChannelIds).Find(&channels).Error
}
if err != nil {
return 0, err
// Process query in chunks to avoid "too many placeholders" error
err = nil
for _, chunk := range lo.Chunk(abilityChannelIds, 100) {
var channelsChunk []Channel
err = DB.Where("id NOT IN (?)", chunk).Find(&channelsChunk).Error
if err != nil {
common.SysError(fmt.Sprintf("Find channels not in abilities table failed: %s", err.Error()))
return count, err
}
channels = append(channels, channelsChunk...)
}
}
for _, channel := range channels {
err := channel.UpdateAbilities(nil)
if err != nil {

View File

@@ -16,6 +16,9 @@ var channelsIDM map[int]*Channel
var channelSyncLock sync.RWMutex
func InitChannelCache() {
if !common.MemoryCacheEnabled {
return
}
newChannelId2channel := make(map[int]*Channel)
var channels []*Channel
DB.Where("status = ?", common.ChannelStatusEnabled).Find(&channels)
@@ -84,11 +87,11 @@ func CacheGetRandomSatisfiedChannel(group string, model string, retry int) (*Cha
if !common.MemoryCacheEnabled {
return GetRandomSatisfiedChannel(group, model, retry)
}
channelSyncLock.RLock()
channels := group2model2channels[group][model]
channelSyncLock.RUnlock()
if len(channels) == 0 {
return nil, errors.New("channel not found")
}

View File

@@ -46,6 +46,17 @@ func (channel *Channel) GetModels() []string {
return strings.Split(strings.Trim(channel.Models, ","), ",")
}
func (channel *Channel) GetGroups() []string {
if channel.Group == "" {
return []string{}
}
groups := strings.Split(strings.Trim(channel.Group, ","), ",")
for i, group := range groups {
groups[i] = strings.TrimSpace(group)
}
return groups
}
func (channel *Channel) GetOtherInfo() map[string]interface{} {
otherInfo := make(map[string]interface{})
if channel.OtherInfo != "" {

View File

@@ -57,6 +57,12 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
if request == nil {
return nil, errors.New("request is nil")
}
// fix: ali parameter.enable_thinking must be set to false for non-streaming calls
if !info.IsStream {
request.EnableThinking = false
}
switch info.RelayMode {
default:
aliReq := requestOpenAI2Ali(*request)

View File

@@ -104,6 +104,65 @@ func DoWssRequest(a Adaptor, c *gin.Context, info *common.RelayInfo, requestBody
return targetConn, nil
}
func startPingKeepAlive(c *gin.Context, pingInterval time.Duration) context.CancelFunc {
pingerCtx, stopPinger := context.WithCancel(context.Background())
gopool.Go(func() {
defer func() {
if common2.DebugEnabled {
println("SSE ping goroutine stopped.")
}
}()
if pingInterval <= 0 {
pingInterval = helper.DefaultPingInterval
}
ticker := time.NewTicker(pingInterval)
// 退出时清理 ticker
defer ticker.Stop()
var pingMutex sync.Mutex
if common2.DebugEnabled {
println("SSE ping goroutine started")
}
for {
select {
// 发送 ping 数据
case <-ticker.C:
if err := sendPingData(c, &pingMutex); err != nil {
return
}
// 收到退出信号
case <-pingerCtx.Done():
return
// request 结束
case <-c.Request.Context().Done():
return
}
}
})
return stopPinger
}
func sendPingData(c *gin.Context, mutex *sync.Mutex) error {
mutex.Lock()
defer mutex.Unlock()
err := helper.PingData(c)
if err != nil {
common2.LogError(c, "SSE ping error: "+err.Error())
return err
}
if common2.DebugEnabled {
println("SSE ping data sent.")
}
return nil
}
func doRequest(c *gin.Context, req *http.Request, info *common.RelayInfo) (*http.Response, error) {
var client *http.Client
var err error
@@ -115,68 +174,28 @@ func doRequest(c *gin.Context, req *http.Request, info *common.RelayInfo) (*http
} else {
client = service.GetHttpClient()
}
// 流式请求 ping 保活
var stopPinger func()
generalSettings := operation_setting.GetGeneralSetting()
pingEnabled := generalSettings.PingIntervalEnabled
var pingerWg sync.WaitGroup
if info.IsStream {
helper.SetEventStreamHeaders(c)
pingInterval := time.Duration(generalSettings.PingIntervalSeconds) * time.Second
var pingerCtx context.Context
pingerCtx, stopPinger = context.WithCancel(c.Request.Context())
if pingEnabled {
pingerWg.Add(1)
gopool.Go(func() {
defer pingerWg.Done()
if pingInterval <= 0 {
pingInterval = helper.DefaultPingInterval
}
ticker := time.NewTicker(pingInterval)
defer ticker.Stop()
var pingMutex sync.Mutex
if common2.DebugEnabled {
println("SSE ping goroutine started")
}
for {
select {
case <-ticker.C:
pingMutex.Lock()
err2 := helper.PingData(c)
pingMutex.Unlock()
if err2 != nil {
common2.LogError(c, "SSE ping error: "+err.Error())
return
}
if common2.DebugEnabled {
println("SSE ping data sent.")
}
case <-pingerCtx.Done():
if common2.DebugEnabled {
println("SSE ping goroutine stopped.")
}
return
}
}
})
// 处理流式请求的 ping 保活
generalSettings := operation_setting.GetGeneralSetting()
if generalSettings.PingIntervalEnabled {
pingInterval := time.Duration(generalSettings.PingIntervalSeconds) * time.Second
stopPinger := startPingKeepAlive(c, pingInterval)
defer stopPinger()
}
}
resp, err := client.Do(req)
// request结束后停止ping
if info.IsStream && pingEnabled {
stopPinger()
pingerWg.Wait()
}
if err != nil {
return nil, err
}
if resp == nil {
return nil, errors.New("resp is nil")
}
_ = req.Body.Close()
_ = c.Request.Body.Close()
return resp, nil

View File

@@ -11,6 +11,8 @@ var awsModelIDMap = map[string]string{
"claude-3-5-sonnet-20241022": "anthropic.claude-3-5-sonnet-20241022-v2:0",
"claude-3-5-haiku-20241022": "anthropic.claude-3-5-haiku-20241022-v1:0",
"claude-3-7-sonnet-20250219": "anthropic.claude-3-7-sonnet-20250219-v1:0",
"claude-sonnet-4-20250514": "anthropic.claude-sonnet-4-20250514-v1:0",
"claude-opus-4-20250514": "anthropic.claude-opus-4-20250514-v1:0",
}
var awsModelCanCrossRegionMap = map[string]map[string]bool{
@@ -41,6 +43,16 @@ var awsModelCanCrossRegionMap = map[string]map[string]bool{
},
"anthropic.claude-3-7-sonnet-20250219-v1:0": {
"us": true,
"ap": true,
"eu": true,
},
"anthropic.claude-sonnet-4-20250514-v1:0": {
"us": true,
"ap": true,
"eu": true,
},
"anthropic.claude-opus-4-20250514-v1:0": {
"us": true,
},
}

View File

@@ -9,6 +9,7 @@ import (
"one-api/relay/channel"
"one-api/relay/channel/openai"
relaycommon "one-api/relay/common"
"strings"
"github.com/gin-gonic/gin"
)
@@ -49,6 +50,18 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
if request == nil {
return nil, errors.New("request is nil")
}
if strings.HasSuffix(info.UpstreamModelName, "-search") {
info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-search")
request.Model = info.UpstreamModelName
toMap := request.ToMap()
toMap["web_search"] = map[string]any{
"enable": true,
"enable_citation": true,
"enable_trace": true,
"enable_status": false,
}
return toMap, nil
}
return request, nil
}

View File

@@ -38,10 +38,10 @@ func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInf
}
func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
if strings.HasPrefix(info.UpstreamModelName, "claude-3") {
a.RequestMode = RequestModeMessage
} else {
if strings.HasPrefix(info.UpstreamModelName, "claude-2") || strings.HasPrefix(info.UpstreamModelName, "claude-instant") {
a.RequestMode = RequestModeCompletion
} else {
a.RequestMode = RequestModeMessage
}
}

View File

@@ -13,6 +13,10 @@ var ModelList = []string{
"claude-3-5-sonnet-20241022",
"claude-3-7-sonnet-20250219",
"claude-3-7-sonnet-20250219-thinking",
"claude-sonnet-4-20250514",
"claude-sonnet-4-20250514-thinking",
"claude-opus-4-20250514",
"claude-opus-4-20250514-thinking",
}
var ChannelName = "claude"

View File

@@ -2,10 +2,10 @@ package gemini
type GeminiChatRequest struct {
Contents []GeminiChatContent `json:"contents"`
SafetySettings []GeminiChatSafetySettings `json:"safety_settings,omitempty"`
GenerationConfig GeminiChatGenerationConfig `json:"generation_config,omitempty"`
SafetySettings []GeminiChatSafetySettings `json:"safetySettings,omitempty"`
GenerationConfig GeminiChatGenerationConfig `json:"generationConfig,omitempty"`
Tools []GeminiChatTool `json:"tools,omitempty"`
SystemInstructions *GeminiChatContent `json:"system_instruction,omitempty"`
SystemInstructions *GeminiChatContent `json:"systemInstruction,omitempty"`
}
type GeminiThinkingConfig struct {
@@ -54,6 +54,7 @@ type GeminiFileData struct {
type GeminiPart struct {
Text string `json:"text,omitempty"`
Thought bool `json:"thought,omitempty"`
InlineData *GeminiInlineData `json:"inlineData,omitempty"`
FunctionCall *FunctionCall `json:"functionCall,omitempty"`
FunctionResponse *FunctionResponse `json:"functionResponse,omitempty"`

View File

@@ -18,6 +18,24 @@ import (
"github.com/gin-gonic/gin"
)
var geminiSupportedMimeTypes = map[string]bool{
"application/pdf": true,
"audio/mpeg": true,
"audio/mp3": true,
"audio/wav": true,
"image/png": true,
"image/jpeg": true,
"text/plain": true,
"video/mov": true,
"video/mpeg": true,
"video/mp4": true,
"video/mpg": true,
"video/avi": true,
"video/wmv": true,
"video/mpegps": true,
"video/flv": true,
}
// Setting safety to the lowest possible values since Gemini is already powerless enough
func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*GeminiChatRequest, error) {
@@ -39,15 +57,22 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
}
if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
if strings.HasSuffix(info.OriginModelName, "-thinking") {
budgetTokens := model_setting.GetGeminiSettings().ThinkingAdapterBudgetTokensPercentage * float64(geminiRequest.GenerationConfig.MaxOutputTokens)
if budgetTokens == 0 || budgetTokens > 24576 {
budgetTokens = 24576
}
geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
ThinkingBudget: common.GetPointer(int(budgetTokens)),
IncludeThoughts: true,
}
if strings.HasSuffix(info.OriginModelName, "-thinking") {
// 如果模型名以 gemini-2.5-pro 开头,不设置 ThinkingBudget
if strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro") {
geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
IncludeThoughts: true,
}
} else {
budgetTokens := model_setting.GetGeminiSettings().ThinkingAdapterBudgetTokensPercentage * float64(geminiRequest.GenerationConfig.MaxOutputTokens)
if budgetTokens == 0 || budgetTokens > 24576 {
budgetTokens = 24576
}
geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
ThinkingBudget: common.GetPointer(int(budgetTokens)),
IncludeThoughts: true,
}
}
} else if strings.HasSuffix(info.OriginModelName, "-nothinking") {
geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
ThinkingBudget: common.GetPointer(0),
@@ -208,14 +233,20 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
}
// 判断是否是url
if strings.HasPrefix(part.GetImageMedia().Url, "http") {
// 是url获取图片的类型和base64编码的数据
// 是url获取文件的类型和base64编码的数据
fileData, err := service.GetFileBase64FromUrl(part.GetImageMedia().Url)
if err != nil {
return nil, fmt.Errorf("get file base64 from url failed: %s", err.Error())
return nil, fmt.Errorf("get file base64 from url '%s' failed: %w", part.GetImageMedia().Url, err)
}
// 校验 MimeType 是否在 Gemini 支持的白名单中
if _, ok := geminiSupportedMimeTypes[strings.ToLower(fileData.MimeType)]; !ok {
return nil, fmt.Errorf("MIME type '%s' from URL '%s' is not supported by Gemini. Supported types are: %v", fileData.MimeType, part.GetImageMedia().Url, getSupportedMimeTypesList())
}
parts = append(parts, GeminiPart{
InlineData: &GeminiInlineData{
MimeType: fileData.MimeType,
MimeType: fileData.MimeType, // 使用原始的 MimeType因为大小写可能对API有意义
Data: fileData.Base64Data,
},
})
@@ -284,100 +315,126 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
return &geminiRequest, nil
}
// Helper function to get a list of supported MIME types for error messages
func getSupportedMimeTypesList() []string {
keys := make([]string, 0, len(geminiSupportedMimeTypes))
for k := range geminiSupportedMimeTypes {
keys = append(keys, k)
}
return keys
}
// cleanFunctionParameters recursively removes unsupported fields from Gemini function parameters.
func cleanFunctionParameters(params interface{}) interface{} {
if params == nil {
return nil
}
paramMap, ok := params.(map[string]interface{})
if !ok {
// Not a map, return as is (e.g., could be an array or primitive)
return params
}
switch v := params.(type) {
case map[string]interface{}:
// Create a copy to avoid modifying the original
cleanedMap := make(map[string]interface{})
for k, val := range v {
cleanedMap[k] = val
}
// Create a copy to avoid modifying the original
cleanedMap := make(map[string]interface{})
for k, v := range paramMap {
cleanedMap[k] = v
}
// Remove unsupported root-level fields
delete(cleanedMap, "default")
delete(cleanedMap, "exclusiveMaximum")
delete(cleanedMap, "exclusiveMinimum")
delete(cleanedMap, "$schema")
delete(cleanedMap, "additionalProperties")
// Remove unsupported root-level fields
delete(cleanedMap, "default")
delete(cleanedMap, "exclusiveMaximum")
delete(cleanedMap, "exclusiveMinimum")
delete(cleanedMap, "$schema")
delete(cleanedMap, "additionalProperties")
// Clean properties
if props, ok := cleanedMap["properties"].(map[string]interface{}); ok && props != nil {
cleanedProps := make(map[string]interface{})
for propName, propValue := range props {
propMap, ok := propValue.(map[string]interface{})
if !ok {
cleanedProps[propName] = propValue // Keep non-map properties
continue
}
// Create a copy of the property map
cleanedPropMap := make(map[string]interface{})
for k, v := range propMap {
cleanedPropMap[k] = v
}
// Remove unsupported fields
delete(cleanedPropMap, "default")
delete(cleanedPropMap, "exclusiveMaximum")
delete(cleanedPropMap, "exclusiveMinimum")
delete(cleanedPropMap, "$schema")
delete(cleanedPropMap, "additionalProperties")
// Check and clean 'format' for string types
if propType, typeExists := cleanedPropMap["type"].(string); typeExists && propType == "string" {
if formatValue, formatExists := cleanedPropMap["format"].(string); formatExists {
if formatValue != "enum" && formatValue != "date-time" {
delete(cleanedPropMap, "format")
}
// Check and clean 'format' for string types
if propType, typeExists := cleanedMap["type"].(string); typeExists && propType == "string" {
if formatValue, formatExists := cleanedMap["format"].(string); formatExists {
if formatValue != "enum" && formatValue != "date-time" {
delete(cleanedMap, "format")
}
}
}
// Recursively clean nested properties within this property if it's an object/array
// Check the type before recursing
if propType, typeExists := cleanedPropMap["type"].(string); typeExists && (propType == "object" || propType == "array") {
cleanedProps[propName] = cleanFunctionParameters(cleanedPropMap)
} else {
cleanedProps[propName] = cleanedPropMap // Assign the cleaned map back if not recursing
// Clean properties
if props, ok := cleanedMap["properties"].(map[string]interface{}); ok && props != nil {
cleanedProps := make(map[string]interface{})
for propName, propValue := range props {
cleanedProps[propName] = cleanFunctionParameters(propValue)
}
cleanedMap["properties"] = cleanedProps
}
cleanedMap["properties"] = cleanedProps
}
// Recursively clean items in arrays if needed (e.g., type: array, items: { ... })
if items, ok := cleanedMap["items"].(map[string]interface{}); ok && items != nil {
cleanedMap["items"] = cleanFunctionParameters(items)
}
// Also handle items if it's an array of schemas
if itemsArray, ok := cleanedMap["items"].([]interface{}); ok {
cleanedItemsArray := make([]interface{}, len(itemsArray))
for i, item := range itemsArray {
cleanedItemsArray[i] = cleanFunctionParameters(item)
// Recursively clean items in arrays
if items, ok := cleanedMap["items"].(map[string]interface{}); ok && items != nil {
cleanedMap["items"] = cleanFunctionParameters(items)
}
cleanedMap["items"] = cleanedItemsArray
}
// Recursively clean other schema composition keywords if necessary
for _, field := range []string{"allOf", "anyOf", "oneOf"} {
if nested, ok := cleanedMap[field].([]interface{}); ok {
cleanedNested := make([]interface{}, len(nested))
for i, item := range nested {
cleanedNested[i] = cleanFunctionParameters(item)
// Also handle items if it's an array of schemas
if itemsArray, ok := cleanedMap["items"].([]interface{}); ok {
cleanedItemsArray := make([]interface{}, len(itemsArray))
for i, item := range itemsArray {
cleanedItemsArray[i] = cleanFunctionParameters(item)
}
cleanedMap[field] = cleanedNested
cleanedMap["items"] = cleanedItemsArray
}
}
return cleanedMap
// Recursively clean other schema composition keywords
for _, field := range []string{"allOf", "anyOf", "oneOf"} {
if nested, ok := cleanedMap[field].([]interface{}); ok {
cleanedNested := make([]interface{}, len(nested))
for i, item := range nested {
cleanedNested[i] = cleanFunctionParameters(item)
}
cleanedMap[field] = cleanedNested
}
}
// Recursively clean patternProperties
if patternProps, ok := cleanedMap["patternProperties"].(map[string]interface{}); ok {
cleanedPatternProps := make(map[string]interface{})
for pattern, schema := range patternProps {
cleanedPatternProps[pattern] = cleanFunctionParameters(schema)
}
cleanedMap["patternProperties"] = cleanedPatternProps
}
// Recursively clean definitions
if definitions, ok := cleanedMap["definitions"].(map[string]interface{}); ok {
cleanedDefinitions := make(map[string]interface{})
for defName, defSchema := range definitions {
cleanedDefinitions[defName] = cleanFunctionParameters(defSchema)
}
cleanedMap["definitions"] = cleanedDefinitions
}
// Recursively clean $defs (newer JSON Schema draft)
if defs, ok := cleanedMap["$defs"].(map[string]interface{}); ok {
cleanedDefs := make(map[string]interface{})
for defName, defSchema := range defs {
cleanedDefs[defName] = cleanFunctionParameters(defSchema)
}
cleanedMap["$defs"] = cleanedDefs
}
// Clean conditional keywords
for _, field := range []string{"if", "then", "else", "not"} {
if nested, ok := cleanedMap[field]; ok {
cleanedMap[field] = cleanFunctionParameters(nested)
}
}
return cleanedMap
case []interface{}:
// Handle arrays of schemas
cleanedArray := make([]interface{}, len(v))
for i, item := range v {
cleanedArray[i] = cleanFunctionParameters(item)
}
return cleanedArray
default:
// Not a map or array, return as is (e.g., could be a primitive)
return params
}
}
func removeAdditionalPropertiesWithDepth(schema interface{}, depth int) interface{} {
@@ -539,6 +596,8 @@ func responseGeminiChat2OpenAI(response *GeminiChatResponse) *dto.OpenAITextResp
if call := getResponseToolCall(&part); call != nil {
toolCalls = append(toolCalls, *call)
}
} else if part.Thought {
choice.Message.ReasoningContent = part.Text
} else {
if part.ExecutableCode != nil {
texts = append(texts, "```"+part.ExecutableCode.Language+"\n"+part.ExecutableCode.Code+"\n```")
@@ -556,7 +615,6 @@ func responseGeminiChat2OpenAI(response *GeminiChatResponse) *dto.OpenAITextResp
choice.Message.SetToolCalls(toolCalls)
isToolCall = true
}
choice.Message.SetStringContent(strings.Join(texts, "\n"))
}
@@ -596,6 +654,7 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
}
var texts []string
isTools := false
isThought := false
if candidate.FinishReason != nil {
// p := GeminiConvertFinishReason(*candidate.FinishReason)
switch *candidate.FinishReason {
@@ -620,6 +679,9 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
call.SetIndex(len(choice.Delta.ToolCalls))
choice.Delta.ToolCalls = append(choice.Delta.ToolCalls, *call)
}
} else if part.Thought {
isThought = true
texts = append(texts, part.Text)
} else {
if part.ExecutableCode != nil {
texts = append(texts, "```"+part.ExecutableCode.Language+"\n"+part.ExecutableCode.Code+"\n```\n")
@@ -632,7 +694,11 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
}
}
}
choice.Delta.SetContentString(strings.Join(texts, "\n"))
if isThought {
choice.Delta.SetReasoningContent(strings.Join(texts, "\n"))
} else {
choice.Delta.SetContentString(strings.Join(texts, "\n"))
}
if isTools {
choice.FinishReason = &constant.FinishReasonToolCalls
}
@@ -716,8 +782,11 @@ func GeminiChatHandler(c *gin.Context, resp *http.Response, info *relaycommon.Re
if err != nil {
return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
}
if common.DebugEnabled {
println(string(responseBody))
}
var geminiResponse GeminiChatResponse
err = json.Unmarshal(responseBody, &geminiResponse)
err = common.DecodeJson(responseBody, &geminiResponse)
if err != nil {
return service.OpenAIErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
}

View File

@@ -273,36 +273,25 @@ func OpenaiHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayI
}
func OpenaiTTSHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {
responseBody, err := io.ReadAll(resp.Body)
if err != nil {
return service.OpenAIErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
}
err = resp.Body.Close()
if err != nil {
return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
}
// Reset response body
resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
// We shouldn't set the header before we parse the response body, because the parse part may fail.
// And then we will have to send an error response, but in this case, the header has already been set.
// So the httpClient will be confused by the response.
// For example, Postman will report error, and we cannot check the response at all.
// the status code has been judged before, if there is a body reading failure,
// it should be regarded as a non-recoverable error, so it should not return err for external retry.
// Analogous to nginx's load balancing, it will only retry if it can't be requested or
// if the upstream returns a specific status code, once the upstream has already written the header,
// the subsequent failure of the response body should be regarded as a non-recoverable error,
// and can be terminated directly.
defer resp.Body.Close()
usage := &dto.Usage{}
usage.PromptTokens = info.PromptTokens
usage.TotalTokens = info.PromptTokens
for k, v := range resp.Header {
c.Writer.Header().Set(k, v[0])
}
c.Writer.WriteHeader(resp.StatusCode)
_, err = io.Copy(c.Writer, resp.Body)
c.Writer.WriteHeaderNow()
_, err := io.Copy(c.Writer, resp.Body)
if err != nil {
return service.OpenAIErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError), nil
common.LogError(c, err.Error())
}
err = resp.Body.Close()
if err != nil {
return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
}
usage := &dto.Usage{}
usage.PromptTokens = info.PromptTokens
usage.TotalTokens = info.PromptTokens
return nil, usage
}

View File

@@ -31,6 +31,8 @@ var claudeModelMap = map[string]string{
"claude-3-5-sonnet-20240620": "claude-3-5-sonnet@20240620",
"claude-3-5-sonnet-20241022": "claude-3-5-sonnet-v2@20241022",
"claude-3-7-sonnet-20250219": "claude-3-7-sonnet@20250219",
"claude-sonnet-4-20250514": "claude-sonnet-4@20250514",
"claude-opus-4-20250514": "claude-opus-4@20250514",
}
const anthropicVersion = "vertex-2023-10-16"
@@ -93,14 +95,23 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
} else {
suffix = "generateContent"
}
return fmt.Sprintf(
"https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/google/models/%s:%s",
region,
adc.ProjectID,
region,
info.UpstreamModelName,
suffix,
), nil
if region == "global" {
return fmt.Sprintf(
"https://aiplatform.googleapis.com/v1/projects/%s/locations/global/publishers/google/models/%s:%s",
adc.ProjectID,
info.UpstreamModelName,
suffix,
), nil
} else {
return fmt.Sprintf(
"https://%s-aiplatform.googleapis.com/v1/projects/%s/locations/%s/publishers/google/models/%s:%s",
region,
adc.ProjectID,
region,
info.UpstreamModelName,
suffix,
), nil
}
} else if a.RequestMode == RequestModeClaude {
if info.IsStream {
suffix = "streamRawPredict?alt=sse"

View File

@@ -41,16 +41,31 @@ func getAndValidImageRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.
imageRequest.Quality = "standard"
}
}
if imageRequest.N == 0 {
imageRequest.N = 1
}
default:
err := common.UnmarshalBodyReusable(c, imageRequest)
if err != nil {
return nil, err
}
if imageRequest.Model == "" {
imageRequest.Model = "dall-e-3"
}
if strings.Contains(imageRequest.Size, "×") {
return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
}
// Not "256x256", "512x512", or "1024x1024"
if imageRequest.Model == "dall-e-2" || imageRequest.Model == "dall-e" {
if imageRequest.Size != "" && imageRequest.Size != "256x256" && imageRequest.Size != "512x512" && imageRequest.Size != "1024x1024" {
return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024 for dall-e-2 or dall-e")
}
if imageRequest.Size == "" {
imageRequest.Size = "1024x1024"
}
} else if imageRequest.Model == "dall-e-3" {
if imageRequest.Size != "" && imageRequest.Size != "1024x1024" && imageRequest.Size != "1024x1792" && imageRequest.Size != "1792x1024" {
return nil, errors.New("size must be one of 1024x1024, 1024x1792 or 1792x1024 for dall-e-3")
@@ -58,74 +73,24 @@ func getAndValidImageRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.
if imageRequest.Quality == "" {
imageRequest.Quality = "standard"
}
// N should between 1 and 10
//if imageRequest.N != 0 && (imageRequest.N < 1 || imageRequest.N > 10) {
// return service.OpenAIErrorWrapper(errors.New("n must be between 1 and 10"), "invalid_field_value", http.StatusBadRequest)
//}
if imageRequest.Size == "" {
imageRequest.Size = "1024x1024"
}
} else if imageRequest.Model == "gpt-image-1" {
if imageRequest.Quality == "" {
imageRequest.Quality = "auto"
}
}
if imageRequest.Prompt == "" {
return nil, errors.New("prompt is required")
}
if imageRequest.N == 0 {
imageRequest.N = 1
}
}
if imageRequest.Prompt == "" {
return nil, errors.New("prompt is required")
}
if imageRequest.Model == "" {
imageRequest.Model = "dall-e-2"
}
if strings.Contains(imageRequest.Size, "×") {
return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
}
if imageRequest.N == 0 {
imageRequest.N = 1
}
if imageRequest.Size == "" {
imageRequest.Size = "1024x1024"
}
err := common.UnmarshalBodyReusable(c, imageRequest)
if err != nil {
return nil, err
}
if imageRequest.Prompt == "" {
return nil, errors.New("prompt is required")
}
if strings.Contains(imageRequest.Size, "×") {
return nil, errors.New("size an unexpected error occurred in the parameter, please use 'x' instead of the multiplication sign '×'")
}
if imageRequest.N == 0 {
imageRequest.N = 1
}
if imageRequest.Size == "" {
imageRequest.Size = "1024x1024"
}
if imageRequest.Model == "" {
imageRequest.Model = "dall-e-2"
}
// x.ai grok-2-image not support size, quality or style
if imageRequest.Size == "empty" {
imageRequest.Size = ""
}
// Not "256x256", "512x512", or "1024x1024"
if imageRequest.Model == "dall-e-2" || imageRequest.Model == "dall-e" {
if imageRequest.Size != "" && imageRequest.Size != "256x256" && imageRequest.Size != "512x512" && imageRequest.Size != "1024x1024" {
return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024, dall-e-3 1024x1792 or 1792x1024")
}
} else if imageRequest.Model == "dall-e-3" {
if imageRequest.Size != "" && imageRequest.Size != "1024x1024" && imageRequest.Size != "1024x1792" && imageRequest.Size != "1792x1024" {
return nil, errors.New("size must be one of 256x256, 512x512, or 1024x1024, dall-e-3 1024x1792 or 1792x1024")
}
if imageRequest.Quality == "" {
imageRequest.Quality = "standard"
}
//if imageRequest.N != 1 {
// return nil, errors.New("n must be 1")
//}
}
// N should between 1 and 10
//if imageRequest.N != 0 && (imageRequest.N < 1 || imageRequest.N > 10) {
// return service.OpenAIErrorWrapper(errors.New("n must be between 1 and 10"), "invalid_field_value", http.StatusBadRequest)
//}
if setting.ShouldCheckPromptSensitive() {
words, err := service.CheckSensitiveInput(imageRequest.Prompt)
if err != nil {
@@ -229,6 +194,10 @@ func ImageHelper(c *gin.Context) *dto.OpenAIErrorWithStatusCode {
requestBody = bytes.NewBuffer(jsonData)
}
if common.DebugEnabled {
println(fmt.Sprintf("image request body: %s", requestBody))
}
statusCodeMappingStr := c.GetString("status_code_mapping")
resp, err := adaptor.DoRequest(c, relayInfo, requestBody)

View File

@@ -32,7 +32,23 @@ func RelayMidjourneyImage(c *gin.Context) {
})
return
}
resp, err := http.Get(midjourneyTask.ImageUrl)
var httpClient *http.Client
if channel, err := model.CacheGetChannel(midjourneyTask.ChannelId); err == nil {
if proxy, ok := channel.GetSetting()["proxy"]; ok {
if proxyURL, ok := proxy.(string); ok && proxyURL != "" {
if httpClient, err = service.NewProxyHttpClient(proxyURL); err != nil {
c.JSON(400, gin.H{
"error": "proxy_url_invalid",
})
return
}
}
}
}
if httpClient == nil {
httpClient = service.GetHttpClient()
}
resp, err := httpClient.Get(midjourneyTask.ImageUrl)
if err != nil {
c.JSON(http.StatusInternalServerError, gin.H{
"error": "http_get_image_failed",

View File

@@ -47,6 +47,20 @@ func getAndValidateTextRequest(c *gin.Context, relayInfo *relaycommon.RelayInfo)
if textRequest.Model == "" {
return nil, errors.New("model is required")
}
if textRequest.WebSearchOptions != nil {
if textRequest.WebSearchOptions.SearchContextSize != "" {
validSizes := map[string]bool{
"high": true,
"medium": true,
"low": true,
}
if !validSizes[textRequest.WebSearchOptions.SearchContextSize] {
return nil, errors.New("invalid search_context_size, must be one of: high, medium, low")
}
} else {
textRequest.WebSearchOptions.SearchContextSize = "medium"
}
}
switch relayInfo.RelayMode {
case relayconstant.RelayModeCompletions:
if textRequest.Prompt == "" {
@@ -76,6 +90,10 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
// get & validate textRequest 获取并验证文本请求
textRequest, err := getAndValidateTextRequest(c, relayInfo)
if textRequest.WebSearchOptions != nil {
c.Set("chat_completion_web_search_context_size", textRequest.WebSearchOptions.SearchContextSize)
}
if err != nil {
common.LogError(c, fmt.Sprintf("getAndValidateTextRequest failed: %s", err.Error()))
return service.OpenAIErrorWrapperLocal(err, "invalid_text_request", http.StatusBadRequest)
@@ -370,9 +388,20 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
Mul(decimal.NewFromInt(int64(webSearchTool.CallCount))).
Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
extraContent += fmt.Sprintf("Web Search 调用 %d 次,上下文大小 %s调用花费 $%s",
extraContent += fmt.Sprintf("Web Search 调用 %d 次,上下文大小 %s调用花费 %s",
webSearchTool.CallCount, webSearchTool.SearchContextSize, dWebSearchQuota.String())
}
} else if strings.HasSuffix(modelName, "search-preview") {
// search-preview 模型不支持 response api
searchContextSize := ctx.GetString("chat_completion_web_search_context_size")
if searchContextSize == "" {
searchContextSize = "medium"
}
webSearchPrice = operation_setting.GetWebSearchPricePerThousand(modelName, searchContextSize)
dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
extraContent += fmt.Sprintf("Web Search 调用 1 次,上下文大小 %s调用花费 %s",
searchContextSize, dWebSearchQuota.String())
}
// file search tool 计费
var dFileSearchQuota decimal.Decimal
@@ -463,10 +492,16 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
other["image_ratio"] = imageRatio
other["image_output"] = imageTokens
}
if !dWebSearchQuota.IsZero() && relayInfo.ResponsesUsageInfo != nil {
if webSearchTool, exists := relayInfo.ResponsesUsageInfo.BuiltInTools[dto.BuildInToolWebSearchPreview]; exists {
if !dWebSearchQuota.IsZero() {
if relayInfo.ResponsesUsageInfo != nil {
if webSearchTool, exists := relayInfo.ResponsesUsageInfo.BuiltInTools[dto.BuildInToolWebSearchPreview]; exists {
other["web_search"] = true
other["web_search_call_count"] = webSearchTool.CallCount
other["web_search_price"] = webSearchPrice
}
} else if strings.HasSuffix(modelName, "search-preview") {
other["web_search"] = true
other["web_search_call_count"] = webSearchTool.CallCount
other["web_search_call_count"] = 1
other["web_search_price"] = webSearchPrice
}
}

View File

@@ -36,6 +36,10 @@ var defaultCacheRatio = map[string]float64{
"claude-3-5-sonnet-20241022": 0.1,
"claude-3-7-sonnet-20250219": 0.1,
"claude-3-7-sonnet-20250219-thinking": 0.1,
"claude-sonnet-4-20250514": 0.1,
"claude-sonnet-4-20250514-thinking": 0.1,
"claude-opus-4-20250514": 0.1,
"claude-opus-4-20250514-thinking": 0.1,
}
var defaultCreateCacheRatio = map[string]float64{
@@ -47,6 +51,10 @@ var defaultCreateCacheRatio = map[string]float64{
"claude-3-5-sonnet-20241022": 1.25,
"claude-3-7-sonnet-20250219": 1.25,
"claude-3-7-sonnet-20250219-thinking": 1.25,
"claude-sonnet-4-20250514": 1.25,
"claude-sonnet-4-20250514-thinking": 1.25,
"claude-opus-4-20250514": 1.25,
"claude-opus-4-20250514-thinking": 1.25,
}
//var defaultCreateCacheRatio = map[string]float64{}

View File

@@ -114,7 +114,9 @@ var defaultModelRatio = map[string]float64{
"claude-3-5-sonnet-20241022": 1.5,
"claude-3-7-sonnet-20250219": 1.5,
"claude-3-7-sonnet-20250219-thinking": 1.5,
"claude-sonnet-4-20250514": 1.5,
"claude-3-opus-20240229": 7.5, // $15 / 1M tokens
"claude-opus-4-20250514": 7.5,
"ERNIE-4.0-8K": 0.120 * RMB,
"ERNIE-3.5-8K": 0.012 * RMB,
"ERNIE-3.5-8K-0205": 0.024 * RMB,
@@ -440,13 +442,15 @@ func getHardcodedCompletionModelRatio(name string) (float64, bool) {
if name == "chatgpt-4o-latest" {
return 3, true
}
if strings.Contains(name, "claude-instant-1") {
return 3, true
} else if strings.Contains(name, "claude-2") {
return 3, true
} else if strings.Contains(name, "claude-3") {
if strings.Contains(name, "claude-3") {
return 5, true
} else if strings.Contains(name, "claude-sonnet-4") || strings.Contains(name, "claude-opus-4") {
return 5, true
} else if strings.Contains(name, "claude-instant-1") || strings.Contains(name, "claude-2") {
return 3, true
}
if strings.HasPrefix(name, "gpt-3.5") {
if name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
// https://openai.com/blog/new-embedding-models-and-api-updates

View File

@@ -871,7 +871,16 @@ const ChannelsTable = () => {
};
const refresh = async () => {
await loadChannels(activePage - 1, pageSize, idSort, enableTagMode);
if (searchKeyword === '' && searchGroup === '' && searchModel === '') {
await loadChannels(activePage - 1, pageSize, idSort, enableTagMode);
} else {
await searchChannels(
searchKeyword,
searchGroup,
searchModel,
enableTagMode,
);
}
};
useEffect(() => {
@@ -879,9 +888,13 @@ const ChannelsTable = () => {
const localIdSort = localStorage.getItem('id-sort') === 'true';
const localPageSize =
parseInt(localStorage.getItem('page-size')) || ITEMS_PER_PAGE;
const localEnableTagMode = localStorage.getItem('enable-tag-mode') === 'true';
const localEnableBatchDelete = localStorage.getItem('enable-batch-delete') === 'true';
setIdSort(localIdSort);
setPageSize(localPageSize);
loadChannels(0, localPageSize, localIdSort, enableTagMode)
setEnableTagMode(localEnableTagMode);
setEnableBatchDelete(localEnableBatchDelete);
loadChannels(0, localPageSize, localIdSort, localEnableTagMode)
.then()
.catch((reason) => {
showError(reason);
@@ -979,8 +992,8 @@ const ChannelsTable = () => {
enableTagMode,
) => {
if (searchKeyword === '' && searchGroup === '' && searchModel === '') {
await loadChannels(0, pageSize, idSort, enableTagMode);
setActivePage(1);
await loadChannels(activePage - 1, pageSize, idSort, enableTagMode);
// setActivePage(1);
return;
}
setSearching(true);
@@ -1477,10 +1490,12 @@ const ChannelsTable = () => {
{t('开启批量操作')}
</Typography.Text>
<Switch
checked={enableBatchDelete}
label={t('开启批量操作')}
uncheckedText={t('关')}
aria-label={t('是否开启批量操作')}
onChange={(v) => {
localStorage.setItem('enable-batch-delete', v + '');
setEnableBatchDelete(v);
}}
/>
@@ -1544,6 +1559,7 @@ const ChannelsTable = () => {
uncheckedText={t('关')}
aria-label={t('是否启用标签聚合')}
onChange={(v) => {
localStorage.setItem('enable-tag-mode', v + '');
setEnableTagMode(v);
loadChannels(0, pageSize, idSort, v);
}}

View File

@@ -39,7 +39,9 @@ const ModelSetting = () => {
item.key === 'claude.default_max_tokens' ||
item.key === 'gemini.supported_imagine_models'
) {
item.value = JSON.stringify(JSON.parse(item.value), null, 2);
if (item.value !== '') {
item.value = JSON.stringify(JSON.parse(item.value), null, 2);
}
}
if (item.key.endsWith('Enabled') || item.key.endsWith('enabled')) {
newInputs[item.key] = item.value === 'true' ? true : false;
@@ -60,6 +62,7 @@ const ModelSetting = () => {
// showSuccess('刷新成功');
} catch (error) {
showError('刷新失败');
console.error(error);
} finally {
setLoading(false);
}

View File

@@ -158,7 +158,7 @@ const Home = () => {
</p>
<p>
{t('OIDC 身份验证')}
{statusState?.status?.oidc === true
{statusState?.status?.oidc_enabled === true
? t('已启用')
: t('未启用')}
</p>

View File

@@ -64,8 +64,9 @@ const Playground = () => {
},
];
const defaultModel = 'gpt-4o-mini';
const [inputs, setInputs] = useState({
model: 'gpt-4o-mini',
model: defaultModel,
group: '',
max_tokens: 0,
temperature: 0,
@@ -108,6 +109,11 @@ const Playground = () => {
value: model,
}));
setModels(localModelOptions);
// if default model is not in the list, set the first one as default
const hasDefault = localModelOptions.some(option => option.value === defaultModel);
if (!hasDefault && localModelOptions.length > 0) {
setInputs((inputs) => ({ ...inputs, model: localModelOptions[0].value }));
}
} else {
showError(t(message));
}

View File

@@ -27,40 +27,48 @@ export default function SettingGeminiModel(props) {
const [inputs, setInputs] = useState({
'gemini.safety_settings': '',
'gemini.version_settings': '',
'gemini.supported_imagine_models': [],
'gemini.supported_imagine_models': '',
'gemini.thinking_adapter_enabled': false,
'gemini.thinking_adapter_budget_tokens_percentage': 0.6,
});
const refForm = useRef();
const [inputsRow, setInputsRow] = useState(inputs);
function onSubmit() {
const updateArray = compareObjects(inputs, inputsRow);
if (!updateArray.length) return showWarning(t('你似乎并没有修改什么'));
const requestQueue = updateArray.map((item) => {
let value = String(inputs[item.key]);
return API.put('/api/option/', {
key: item.key,
value,
});
});
setLoading(true);
Promise.all(requestQueue)
.then((res) => {
if (requestQueue.length === 1) {
if (res.includes(undefined)) return;
} else if (requestQueue.length > 1) {
if (res.includes(undefined))
return showError(t('部分保存失败,请重试'));
}
showSuccess(t('保存成功'));
props.refresh();
async function onSubmit() {
await refForm.current
.validate()
.then(() => {
const updateArray = compareObjects(inputs, inputsRow);
if (!updateArray.length) return showWarning(t('你似乎并没有修改什么'));
const requestQueue = updateArray.map((item) => {
let value = String(inputs[item.key]);
return API.put('/api/option/', {
key: item.key,
value,
});
});
setLoading(true);
Promise.all(requestQueue)
.then((res) => {
if (requestQueue.length === 1) {
if (res.includes(undefined)) return;
} else if (requestQueue.length > 1) {
if (res.includes(undefined))
return showError(t('部分保存失败,请重试'));
}
showSuccess(t('保存成功'));
props.refresh();
})
.catch(() => {
showError(t('保存失败,请重试'));
})
.finally(() => {
setLoading(false);
});
})
.catch(() => {
showError(t('保存失败,请重试'));
})
.finally(() => {
setLoading(false);
.catch((error) => {
console.error('Validation failed:', error);
showError(t('请检查输入'));
});
}
@@ -146,6 +154,14 @@ export default function SettingGeminiModel(props) {
label={t('支持的图像模型')}
placeholder={t('例如:') + '\n' + JSON.stringify(['gemini-2.0-flash-exp-image-generation'], null, 2)}
onChange={(value) => setInputs({ ...inputs, 'gemini.supported_imagine_models': value })}
trigger='blur'
stopValidateWithError
rules={[
{
validator: (rule, value) => verifyJSON(value),
message: t('不是合法的 JSON 字符串'),
},
]}
/>
</Col>
</Row>