add 3 params

app.py

@@ -2,7 +2,17 @@ import gradio as gr
 from openai import OpenAI
 
 
-def predict(message, history, system_prompt, model, api_url, api_key):
+def predict(
+    message,
+    history,
+    system_prompt,
+    model,
+    api_url,
+    api_key,
+    max_tk,
+    temp,
+    top_p,
+):
     # Format history with a given chat template
     msgs = [{"role": "system", "content": system_prompt}]
     for user, assistant in history:
@@ -15,7 +25,9 @@ def predict(message, history, system_prompt, model, api_url, api_key):
         response = client.chat.completions.create(
             model=model,
             messages=msgs,
-            temperature=0.3,
+            max_tokens=max_tk,
+            temperature=temp,
+            top_p=top_p,
             stream=False,
         ).to_dict()["choices"][0]["message"]["content"]
 
@@ -25,7 +37,16 @@ def predict(message, history, system_prompt, model, api_url, api_key):
     return response
 
 
-def deepseek(message, history, model, api_key, system_prompt):
+def deepseek(
+    message,
+    history,
+    model,
+    api_key,
+    system_prompt,
+    max_tk,
+    temp,
+    top_p,
+):
     response = predict(
         message,
         history,
@@ -33,6 +54,9 @@ def deepseek(message, history, model, api_key, system_prompt):
         model,
         "https://api.deepseek.com",
         api_key,
+        max_tk,
+        temp,
+        top_p,
     )
     outputs = []
     for new_token in response:
@@ -40,7 +64,16 @@ def deepseek(message, history, model, api_key, system_prompt):
         yield "".join(outputs)
 
 
-def kimi(message, history, model, api_key, system_prompt):
+def kimi(
+    message,
+    history,
+    model,
+    api_key,
+    system_prompt,
+    max_tk,
+    temp,
+    top_p,
+):
     response = predict(
         message,
         history,
@@ -48,6 +81,9 @@ def kimi(message, history, model, api_key, system_prompt):
         model,
         "https://api.moonshot.cn/v1",
         api_key,
+        max_tk,
+        temp,
+        top_p,
     )
     outputs = []
     for new_token in response:
@@ -74,10 +110,20 @@ if __name__ == "__main__":
                     "You are a useful assistant. first recognize user request and then reply carfuly and thinking",
                     label="System prompt",
                 )
+                ds_maxtk = gr.Slider(0, 32000, 10000, label="Max new tokens")
+                ds_temp = gr.Slider(0, 1, 0.3, label="Temperature")
+                ds_topp = gr.Slider(0, 1, 0.95, label="Top P sampling")
 
             gr.ChatInterface(
                 deepseek,
-                additional_inputs=[ds_model, ds_key, ds_sys],
+                additional_inputs=[
+                    ds_model,
+                    ds_key,
+                    ds_sys,
+                    ds_maxtk,
+                    ds_temp,
+                    ds_topp,
+                ],
             )
 
         with gr.Tab("Kimi"):
@@ -96,10 +142,20 @@ if __name__ == "__main__":
                     "You are a useful assistant. first recognize user request and then reply carfuly and thinking",
                     label="System prompt",
                 )
+                kimi_maxtk = gr.Slider(0, 32000, 10000, label="Max new tokens")
+                kimi_temp = gr.Slider(0, 1, 0.3, label="Temperature")
+                kimi_topp = gr.Slider(0, 1, 0.95, label="Top P sampling")
 
             gr.ChatInterface(
                 kimi,
-                additional_inputs=[kimi_model, kimi_key, kimi_sys],
+                additional_inputs=[
+                    kimi_model,
+                    kimi_key,
+                    kimi_sys,
+                    kimi_maxtk,
+                    kimi_temp,
+                    kimi_topp,
+                ],
             )
 
     demo.queue().launch()