refactor(api): update temperature and penalty parameters for DeepSeek…

… and OpenAI models

- Made temperature, presence_penalty, frequency_penalty, and top_p optional in RequestPayload interface.
- Adjusted request payload construction in DeepSeekApi to conditionally include temperature for the "deepseek-reasoner" model.
- Updated ChatGPTApi to handle temperature and penalty parameters based on model type, specifically for "deepseek-reasoner" and "o1" models.
- Ensured compatibility with existing models while enhancing flexibility for new configurations.

app/client/platforms/deepseek.ts

@@ -84,14 +84,15 @@ export class DeepSeekApi implements LLMApi {
       messages,
       stream: options.config.stream,
       model: modelConfig.model,
-      temperature: modelConfig.temperature,
       presence_penalty: modelConfig.presence_penalty,
       frequency_penalty: modelConfig.frequency_penalty,
       top_p: modelConfig.top_p,
       // max_tokens: Math.max(modelConfig.max_tokens, 1024),
       // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
     };
-
+    if (modelConfig.model === "deepseek-reasoner") {
+      requestPayload.temperature = modelConfig.temperature;
+    }
     console.log("[Request] openai payload: ", requestPayload);
 
     const shouldStream = !!options.config.stream;

app/client/platforms/openai.ts

@@ -60,10 +60,10 @@ export interface RequestPayload {
   }[];
   stream?: boolean;
   model: string;
-  temperature: number;
-  presence_penalty: number;
-  frequency_penalty: number;
-  top_p: number;
+  temperature?: number;
+  presence_penalty?: number;
+  frequency_penalty?: number;
+  top_p?: number;
   max_tokens?: number;
   max_completion_tokens?: number;
 }
@@ -196,6 +196,8 @@ export class ChatGPTApi implements LLMApi {
 
     const isDalle3 = _isDalle3(options.config.model);
     const isO1 = options.config.model.startsWith("o1");
+    const isDeepseekReasoner =
+      options.config.model.startsWith("deepseek-reasoner");
     if (isDalle3) {
       const prompt = getMessageTextContent(
         options.messages.slice(-1)?.pop() as any,
@@ -226,16 +228,25 @@ export class ChatGPTApi implements LLMApi {
         messages,
         stream: options.config.stream,
         model: modelConfig.model,
-        temperature: !isO1 ? modelConfig.temperature : 1,
-        presence_penalty: !isO1 ? modelConfig.presence_penalty : 0,
-        frequency_penalty: !isO1 ? modelConfig.frequency_penalty : 0,
-        top_p: !isO1 ? modelConfig.top_p : 1,
+        temperature:
+          !isO1 && !isDeepseekReasoner ? modelConfig.temperature : undefined,
+        presence_penalty: !isDeepseekReasoner
+          ? isO1
+            ? 0
+            : modelConfig.presence_penalty
+          : undefined,
+        frequency_penalty: !isDeepseekReasoner
+          ? isO1
+            ? 0
+            : modelConfig.frequency_penalty
+          : undefined,
+        top_p: !isDeepseekReasoner ? (isO1 ? 1 : modelConfig.top_p) : undefined,
         // max_tokens: Math.max(modelConfig.max_tokens, 1024),
         // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
       };
 
-      // O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
-      if (isO1) {
+      // O1 使用 max_completion_tokens 控制token数
+      if (isO1 || isDeepseekReasoner) {
         requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
       }
 

src-tauri/tauri.conf.json

@@ -9,7 +9,7 @@
   },
   "package": {
     "productName": "NextChat",
-    "version": "2.15.9"
+    "version": "2.15.10-beta.1"
   },
   "tauri": {
     "allowlist": {