Merge dabaa795b9 into 8b8cd03e85

* feat: support gemini-2.0-flash

- Enhance model support by adding new entries and refining checks for system instruction compatibility.
- Update logging display behavior and adjust default quotas for better user experience.
- Revamp pricing structures in the billing system to reflect current model values and deprecate outdated entries.
- Streamline code by replacing hardcoded values with configurations for maintainability.

* feat: add new Gemini 2.0 flash models to adapter and billing ratio

* fix: update GetRequestURL to support gemini-1.5 model in versioning

README.md

@@ -115,7 +115,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
 19. 支持丰富的**自定义**设置，
     1. 支持自定义系统名称，logo 以及页脚。
     2. 支持自定义首页和关于页面，可以选择使用 HTML & Markdown 代码进行自定义，或者使用一个单独的网页通过 iframe 嵌入。
-20. 支持通过系统访问令牌调用管理 API，进而**在无需二开的情况下扩展和自定义** One API 的功能，详情请参考此处 [API 文档](./docs/API.md)。。
+20. 支持通过系统访问令牌调用管理 API，进而**在无需二开的情况下扩展和自定义** One API 的功能，详情请参考此处 [API 文档](./docs/API.md)。
 21. 支持 Cloudflare Turnstile 用户校验。
 22. 支持用户管理，支持**多种用户登录注册方式**：
     + 邮箱登录注册（支持注册邮箱白名单）以及通过邮箱进行密码重置。

relay/adaptor/gemini/adaptor.go

@@ -5,9 +5,10 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"strings"
 
 	"github.com/gin-gonic/gin"
-
+	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
 	channelhelper "github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
@@ -20,17 +21,12 @@ type Adaptor struct {
 }
 
 func (a *Adaptor) Init(meta *meta.Meta) {
-
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
-	var defaultVersion string
-	switch meta.ActualModelName {
-	case "gemini-2.0-flash-exp",
-		"gemini-2.0-flash-thinking-exp",
-		"gemini-2.0-flash-thinking-exp-01-21":
-		defaultVersion = "v1beta"
-	default:
+	defaultVersion := config.GeminiVersion
+	if strings.Contains(meta.ActualModelName, "gemini-2.0") ||
+		strings.Contains(meta.ActualModelName, "gemini-1.5") {
 		defaultVersion = "v1beta"
 	}
 

relay/adaptor/gemini/constants.go

@@ -4,8 +4,38 @@ package gemini
 
 var ModelList = []string{
 	"gemini-pro", "gemini-1.0-pro",
-	"gemini-1.5-flash", "gemini-1.5-pro",
+	// "gemma-2-2b-it", "gemma-2-9b-it", "gemma-2-27b-it",
+	"gemini-1.5-flash", "gemini-1.5-flash-8b",
+	"gemini-1.5-pro", "gemini-1.5-pro-experimental",
 	"text-embedding-004", "aqa",
-	"gemini-2.0-flash-exp",
-	"gemini-2.0-flash-thinking-exp", "gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-flash", "gemini-2.0-flash-exp",
+	"gemini-2.0-flash-lite-preview-02-05",
+	"gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-pro-exp-02-05",
+}
+
+// ModelsSupportSystemInstruction is the list of models that support system instruction.
+//
+// https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/system-instructions
+var ModelsSupportSystemInstruction = []string{
+	// "gemini-1.0-pro-002",
+	// "gemini-1.5-flash", "gemini-1.5-flash-001", "gemini-1.5-flash-002",
+	// "gemini-1.5-flash-8b",
+	// "gemini-1.5-pro", "gemini-1.5-pro-001", "gemini-1.5-pro-002",
+	// "gemini-1.5-pro-experimental",
+	"gemini-2.0-flash", "gemini-2.0-flash-exp",
+	"gemini-2.0-flash-thinking-exp-01-21",
+}
+
+// IsModelSupportSystemInstruction check if the model support system instruction.
+//
+// Because the main version of Go is 1.20, slice.Contains cannot be used
+func IsModelSupportSystemInstruction(model string) bool {
+	for _, m := range ModelsSupportSystemInstruction {
+		if m == model {
+			return true
+		}
+	}
+
+	return false
 }

relay/adaptor/gemini/main.go

@@ -132,9 +132,16 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 		}
 		// Converting system prompt to prompt from user for the same reason
 		if content.Role == "system" {
-			content.Role = "user"
 			shouldAddDummyModelMessage = true
+			if IsModelSupportSystemInstruction(textRequest.Model) {
+				geminiRequest.SystemInstruction = &content
+				geminiRequest.SystemInstruction.Role = ""
+				continue
+			} else {
+				content.Role = "user"
+			}
 		}
+
 		geminiRequest.Contents = append(geminiRequest.Contents, content)
 
 		// If a system message is the last message, we need to add a dummy model message to make gemini happy

relay/adaptor/gemini/model.go

@@ -1,10 +1,11 @@
 package gemini
 
 type ChatRequest struct {
-	Contents         []ChatContent        `json:"contents"`
-	SafetySettings   []ChatSafetySettings `json:"safety_settings,omitempty"`
-	GenerationConfig ChatGenerationConfig `json:"generation_config,omitempty"`
-	Tools            []ChatTools          `json:"tools,omitempty"`
+	Contents          []ChatContent        `json:"contents"`
+	SafetySettings    []ChatSafetySettings `json:"safety_settings,omitempty"`
+	GenerationConfig  ChatGenerationConfig `json:"generation_config,omitempty"`
+	Tools             []ChatTools          `json:"tools,omitempty"`
+	SystemInstruction *ChatContent         `json:"system_instruction,omitempty"`
 }
 
 type EmbeddingRequest struct {

relay/adaptor/openai/adaptor.go

@@ -32,16 +32,24 @@ func (a *Adaptor) Init(meta *meta.Meta) {
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	switch meta.ChannelType {
 	case channeltype.Azure:
+		defaultVersion := meta.Config.APIVersion
+
+		// https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning?tabs=python#api--feature-support
+		if strings.HasPrefix(meta.ActualModelName, "o1") ||
+			strings.HasPrefix(meta.ActualModelName, "o3") {
+			defaultVersion = "2024-12-01-preview"
+		}
+
 		if meta.Mode == relaymode.ImagesGenerations {
 			// https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart?tabs=dalle3%2Ccommand-line&pivots=rest-api
 			// https://{resource_name}.openai.azure.com/openai/deployments/dall-e-3/images/generations?api-version=2024-03-01-preview
-			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, meta.Config.APIVersion)
+			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, defaultVersion)
 			return fullRequestURL, nil
 		}
 
 		// https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
 		requestURL := strings.Split(meta.RequestURLPath, "?")[0]
-		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, meta.Config.APIVersion)
+		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, defaultVersion)
 		task := strings.TrimPrefix(requestURL, "/v1/")
 		model_ := meta.ActualModelName
 		model_ = strings.Replace(model_, ".", "", -1)
@@ -89,6 +97,23 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		}
 		request.StreamOptions.IncludeUsage = true
 	}
+
+	// o1/o1-mini/o1-preview do not support system prompt/max_tokens/temperature
+	if strings.HasPrefix(request.Model, "o1") {
+		temperature := float64(1)
+		request.Temperature = &temperature // Only the default (1) value is supported
+		request.MaxTokens = 0
+		request.Messages = func(raw []model.Message) (filtered []model.Message) {
+			for i := range raw {
+				if raw[i].Role != "system" {
+					filtered = append(filtered, raw[i])
+				}
+			}
+
+			return
+		}(request.Messages)
+	}
+
 	return request, nil
 }
 

relay/adaptor/vertexai/gemini/adapter.go

@@ -16,10 +16,12 @@ import (
 
 var ModelList = []string{
 	"gemini-pro", "gemini-pro-vision",
-	"gemini-1.5-pro-001", "gemini-1.5-flash-001",
-	"gemini-1.5-pro-002", "gemini-1.5-flash-002",
-	"gemini-2.0-flash-exp",
-	"gemini-2.0-flash-thinking-exp", "gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-exp-1206",
+	"gemini-1.5-pro-001", "gemini-1.5-pro-002",
+	"gemini-1.5-flash-001", "gemini-1.5-flash-002",
+	"gemini-2.0-flash-exp", "gemini-2.0-flash-001",
+	"gemini-2.0-flash-lite-preview-02-05",
+	"gemini-2.0-flash-thinking-exp-01-21",
 }
 
 type Adaptor struct {

relay/billing/ratio/model.go

@@ -115,15 +115,24 @@ var ModelRatio = map[string]float64{
 	"bge-large-en":       0.002 * RMB,
 	"tao-8k":             0.002 * RMB,
 	// https://ai.google.dev/pricing
-	"gemini-pro":                          1, // $0.00025 / 1k characters -> $0.001 / 1k tokens
-	"gemini-1.0-pro":                      1,
-	"gemini-1.5-pro":                      1,
-	"gemini-1.5-pro-001":                  1,
-	"gemini-1.5-flash":                    1,
-	"gemini-1.5-flash-001":                1,
-	"gemini-2.0-flash-exp":                1,
-	"gemini-2.0-flash-thinking-exp":       1,
-	"gemini-2.0-flash-thinking-exp-01-21": 1,
+	// https://cloud.google.com/vertex-ai/generative-ai/pricing
+	// "gemma-2-2b-it":                       0,
+	// "gemma-2-9b-it":                       0,
+	// "gemma-2-27b-it":                      0,
+	"gemini-pro":                          0.25 * MILLI_USD, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	"gemini-1.0-pro":                      0.125 * MILLI_USD,
+	"gemini-1.5-pro":                      1.25 * MILLI_USD,
+	"gemini-1.5-pro-001":                  1.25 * MILLI_USD,
+	"gemini-1.5-pro-experimental":         1.25 * MILLI_USD,
+	"gemini-1.5-flash":                    0.075 * MILLI_USD,
+	"gemini-1.5-flash-001":                0.075 * MILLI_USD,
+	"gemini-1.5-flash-8b":                 0.0375 * MILLI_USD,
+	"gemini-2.0-flash-exp":                0.075 * MILLI_USD,
+	"gemini-2.0-flash":                    0.15 * MILLI_USD,
+	"gemini-2.0-flash-001":                0.15 * MILLI_USD,
+	"gemini-2.0-flash-lite-preview-02-05": 0.075 * MILLI_USD,
+	"gemini-2.0-flash-thinking-exp-01-21": 0.075 * MILLI_USD,
+	"gemini-2.0-pro-exp-02-05":            1.25 * MILLI_USD,
 	"aqa":                                 1,
 	// https://open.bigmodel.cn/pricing
 	"glm-zero-preview": 0.01 * RMB,
@@ -506,7 +515,7 @@ func GetCompletionRatio(name string, channelType int) float64 {
 		}
 		return 2
 	}
-	// including o1, o1-preview, o1-mini
+	// including o1/o1-preview/o1-mini
 	if strings.HasPrefix(name, "o1") {
 		return 4
 	}

web/default/src/components/ChannelsTable.js

@@ -44,6 +44,9 @@ function renderType(type, t) {
 function renderBalance(type, balance, t) {
   switch (type) {
     case 1: // OpenAI
+        if (balance === 0) {
+            return <span>{t('channel.table.balance_not_supported')}</span>;
+        }
       return <span>${balance.toFixed(2)}</span>;
     case 4: // CloseAI
       return <span>¥{balance.toFixed(2)}</span>;
@@ -108,7 +111,7 @@ const ChannelsTable = () => {
 
   const loadChannels = async (startIdx) => {
     const res = await API.get(`/api/channel/?p=${startIdx}`);
-    const {success, message, data} = res.data;
+    const { success, message, data } = res.data;
     if (success) {
       let localChannels = data.map(processChannelData);
       if (startIdx === 0) {
@@ -588,7 +591,15 @@ const ChannelsTable = () => {
                     />
                   </Table.Cell>
                   <Table.Cell>
-                    <div>
+                    <div
+                      style={{
+                        display: 'flex',
+                        alignItems: 'center',
+                        flexWrap: 'wrap',
+                        gap: '2px',
+                        rowGap: '6px',
+                      }}
+                    >
                       <Button
                         size={'tiny'}
                         positive

web/default/src/helpers/render.js

@@ -1,5 +1,5 @@
-import {Label, Message} from 'semantic-ui-react';
-import {getChannelOption} from './helper';
+import { Label, Message } from 'semantic-ui-react';
+import { getChannelOption } from './helper';
 import React from 'react';
 
 export function renderText(text, limit) {
@@ -16,7 +16,15 @@ export function renderGroup(group) {
   let groups = group.split(',');
   groups.sort();
   return (
-    <>
+    <div
+      style={{
+        display: 'flex',
+        alignItems: 'center',
+        flexWrap: 'wrap',
+        gap: '2px',
+        rowGap: '6px',
+      }}
+    >
       {groups.map((group) => {
         if (group === 'vip' || group === 'pro') {
           return <Label color='yellow'>{group}</Label>;
@@ -25,7 +33,7 @@ export function renderGroup(group) {
         }
         return <Label>{group}</Label>;
       })}
-    </>
+    </div>
   );
 }
 
@@ -106,8 +114,8 @@ export function renderChannelTip(channelId) {
     return <></>;
   }
   return (
-      <Message>
-        <div dangerouslySetInnerHTML={{__html: channel.tip}}></div>
-      </Message>
+    <Message>
+      <div dangerouslySetInnerHTML={{ __html: channel.tip }}></div>
+    </Message>
   );
 }

web/default/src/helpers/utils.js

@@ -1,7 +1,7 @@
-import { toast } from 'react-toastify';
-import { toastConstants } from '../constants';
+import {toast} from 'react-toastify';
+import {toastConstants} from '../constants';
 import React from 'react';
-import { API } from './api';
+import {API} from './api';
 
 const HTMLToastContent = ({ htmlContent }) => {
   return <div dangerouslySetInnerHTML={{ __html: htmlContent }} />;
@@ -74,6 +74,7 @@ if (isMobile()) {
 }
 
 export function showError(error) {
+  if (!error) return;
   console.error(error);
   if (error.message) {
     if (error.name === 'AxiosError') {
@@ -158,17 +159,7 @@ export function timestamp2string(timestamp) {
     second = '0' + second;
   }
   return (
-    year +
-    '-' +
-    month +
-    '-' +
-    day +
-    ' ' +
-    hour +
-    ':' +
-    minute +
-    ':' +
-    second
+      year + '-' + month + '-' + day + ' ' + hour + ':' + minute + ':' + second
   );
 }
 
@@ -193,7 +184,6 @@ export const verifyJSON = (str) => {
 export function shouldShowPrompt(id) {
   let prompt = localStorage.getItem(`prompt-${id}`);
   return !prompt;
-
 }
 
 export function setPromptShown(id) {
@@ -224,4 +214,4 @@ export function getChannelModels(type) {
     return channelModels[type];
   }
   return [];
-}
+}