v0.6.2.4 开发分支，支持 openai max token 设置, 调整插件依赖

tooweakchen · May 27, 2023 · 020341c · 020341c
1 parent 3c78964
commit 020341c
Show file tree

Hide file tree

Showing 7 changed files with 31 additions and 10 deletions.
diff --git a/chat/common/openai/chat.go b/chat/common/openai/chat.go
@@ -29,7 +29,7 @@ func (c *ChatClient) ChatStream(req []ChatModelMessage, channel chan string) (st
 			continue
 		}
 		//估算长度
-		if NumTokensFromMessages(req[len(req)-i-1:], ChatModel) < (TotalToken - c.MaxToken) {
+		if NumTokensFromMessages(req[len(req)-i-1:], ChatModel) < (c.TotalToken - c.MaxToken) {
 			first = len(req) - i - 1
 		} else {
 			break
@@ -51,6 +51,7 @@ func (c *ChatClient) ChatStream(req []ChatModelMessage, channel chan string) (st
 			Content: message.Content,
 		})
 	}
+	fmt.Println("current", c.Model)
 	if _, ok := Models[c.Model]; !ok {
 		c.Model = ChatModel
 	}
@@ -117,7 +118,7 @@ func (c *ChatClient) Chat(req []ChatModelMessage) (string, error) {
 			continue
 		}
 		//估算长度
-		if NumTokensFromMessages(req[len(req)-i-1:], ChatModel) < (TotalToken - c.MaxToken) {
+		if NumTokensFromMessages(req[len(req)-i-1:], ChatModel) < (c.TotalToken - c.MaxToken) {
 			first = len(req) - i - 1
 		} else {
 			break

diff --git a/chat/common/openai/context.go b/chat/common/openai/context.go
@@ -197,7 +197,8 @@ func (c *UserContext) doSummary(summary []ChatModelMessage) ([]ChatModelMessage,
 	logx.Info("summary_req_length", ": ", len([]rune(newPrompt)))
 
 	// 调用 openai api 进行 summary 简化到 100 字以内
-	summaryStr, err := c.Client.WithModel(TextModel).WithMaxToken(1500).WithTemperature(0.1).
+	sc := c
+	summaryStr, err := sc.Client.WithModel(TextModel).WithMaxToken(1500).WithTemperature(0).
 		Completion(newPrompt)
 
 	logx.Info("summary_reps", ": "+summaryStr)

diff --git a/chat/common/openai/openai.go b/chat/common/openai/openai.go
@@ -51,13 +51,15 @@ type ChatClient struct {
 	Model       string  `json:"model"`
 	BaseHost    string  `json:"base_host"`
 	MaxToken    int     `json:"max_token"`
+	TotalToken  int     `json:"total_token"`
 	Temperature float32 `json:"temperature"`
 }
 
 func NewChatClient(apiKey string) *ChatClient {
 	return &ChatClient{
 		APIKey:      apiKey,
 		MaxToken:    MaxToken,
+		TotalToken:  TotalToken,
 		Temperature: float32(Temperature),
 	}
 }
@@ -98,6 +100,12 @@ func (c *ChatClient) WithTemperature(temperature float32) *ChatClient {
 	return c
 }
 
+// WithTotalToken 设置总token数
+func (c *ChatClient) WithTotalToken(totalToken int) *ChatClient {
+	c.TotalToken = totalToken
+	return c
+}
+
 func (c *ChatClient) WithHttpProxy(proxyUrl string) *ChatClient {
 	c.HttpProxy = proxyUrl
 	return c

diff --git a/chat/service/chat/api/etc/chat-api.yaml.complete.bak b/chat/service/chat/api/etc/chat-api.yaml.complete.bak
@@ -35,6 +35,9 @@ OpenAi:                                             # openai配置
   Host: "https://api.openai.com"                    # openai host （可选，使用cf进行反向代理时，修改可用）
   Origin: "open_ai"                                 # 默认为 调用 open_ai 也支持 azure , azure_ad (可选 默认为 open_ai)
   Engine: "deployment_name"                         # engine = "deployment_name"(当 Origin 为 azure, azure_ad  时必填)
+  MaxToken: 2000                                    # 一次会话能响应内容的最大 token
+  TotalToken 3900                                   # 一次对话话 openai 能处理的最大 token 数量 gpt3:4096 /gpt4:8192 /gpt-4-32k:32768
+  Temperature: 0.8                                  # 对话的创造性，当其逼近与0时，其响应的结果更加死板，当其趋近于1时，其对话更加符有跳跃与创造力
 
 Proxy:                                              # 代理配置 （可选）
   Enable: false                                     # 是否启用代理，默认为 false（可选）
@@ -75,7 +78,7 @@ Plugins:
       Auth:
         Type: "none"
       API:
-        URL: "http://192.168.1.202:8885/chat"
+        URL: "http://192.168.1.202:8885/search"
     - NameForHuman: "维基百科查询"
       NameForModel: "wikipedia"
       DescForHuman: "这个插件可以提供关于人、地点、公司、历史事件或其他主题的一般性问题。"

diff --git a/chat/service/chat/api/internal/config/config.go b/chat/service/chat/api/internal/config/config.go
@@ -48,10 +48,13 @@ type Config struct {
 
 	// openai 配置
 	OpenAi struct {
-		Key    string
-		Host   string `json:"host,optional,default=https://api.openai.com"`
-		Origin string `json:"origin,optional,default=open_ai"`
-		Engine string `json:"engine,optional,default="`
+		Key         string
+		Host        string  `json:"host,optional,default=https://api.openai.com"`
+		Origin      string  `json:"origin,optional,default=open_ai"`
+		Engine      string  `json:"engine,optional,default="`
+		MaxToken    int     `json:"max_token,optional,default=2000"`
+		TotalToken  int     `json:"total_token,optional,default=3900"`
+		Temperature float32 `json:"temperature,optional,default=0.8"`
 	}
 
 	// http proxy 设置

diff --git a/chat/service/chat/api/internal/logic/chatlogic.go b/chat/service/chat/api/internal/logic/chatlogic.go
@@ -78,7 +78,11 @@ func (l *ChatLogic) Chat(req *types.ChatReq) (resp *types.ChatReply, err error)
 			WithModel(l.model).
 			WithBaseHost(l.baseHost).
 			WithOrigin(l.svcCtx.Config.OpenAi.Origin).
-			WithEngine(l.svcCtx.Config.OpenAi.Engine)
+			WithEngine(l.svcCtx.Config.OpenAi.Engine).
+			WithMaxToken(l.svcCtx.Config.OpenAi.MaxToken).
+			WithTemperature(l.svcCtx.Config.OpenAi.Temperature).
+			WithTotalToken(l.svcCtx.Config.OpenAi.TotalToken)
+
 		if l.svcCtx.Config.Proxy.Enable {
 			c = c.WithHttpProxy(l.svcCtx.Config.Proxy.Http).WithSocks5Proxy(l.svcCtx.Config.Proxy.Socket5)
 		}
@@ -185,7 +189,8 @@ func (l *ChatLogic) Chat(req *types.ChatReq) (resp *types.ChatReply, err error)
 						API:          i2.API,
 					})
 				}
-				pluginInfo, err := c.WithMaxToken(1000).WithTemperature(0).
+				pc := c
+				pluginInfo, err := pc.WithMaxToken(1000).WithTemperature(0).
 					Completion(plugin.GetPluginPromptInfo(req.MSG, p))
 				if err == nil {
 					msg, ok := plugin.RunPlugin(pluginInfo, p)

diff --git a/plugins/search/requirements.txt b/plugins/search/requirements.txt