Update app.py

max_tokens 16k for deepseek models and msg if finish_reason == 'length'

app.py

@@ -676,6 +676,7 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
         using_groq = False
         reasoning = False
         prompt = prompt.strip()
+        finish_reason = 'ok'
         if prompt.lower().startswith('dsr1 '):
             deepseek = True
             ds_model = 'deepseek-ai/DeepSeek-R1'
@@ -723,15 +724,20 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
                     completion = client.chat.completions.create(
                         temperature=0.6,
                         model= ds_model,
-                        messages=past)
+                        messages=past,
+                        )
                     reporting_model='deepseek70-groq'
                 else:
                     client = OpenAI(api_key=DEEPSEEK_KEY, base_url='https://api.together.xyz/v1')
                     completion = client.chat.completions.create(
                         temperature=0.6,
                         model= ds_model,
-                        messages=past)
+                        messages=past,
+                        max_tokens=16000
+                        )
                     reporting_model='deepseek-together-' + ds_model[-3:].replace('.5B','1.5B')
+                    if completion.choices[0].finish_reason == 'length':
+                        finish_reason = "Truncated due to token limit"
             else:
                 completion = Client().chat.completions.create(model=gptModel,
                                             messages=past)
@@ -758,9 +764,11 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
             tokens_in = completion.usage.prompt_tokens
             tokens_out = completion.usage.completion_tokens
             tokens = completion.usage.total_tokens
-        response += md("=================\nYOU:\n" + prompt + "\n=================\nGPT:\n" + reply)
+        response += md("\n\n***YOU***: " + prompt + "\n\n***GPT***: ") +  reply.replace('```','\n\n```\n\n')
         if isBoss:
-            response += md(f"\n{reporting_model}: tokens in/out = {tokens_in}/{tokens_out}\n")
+            response += md(f"\n\n{reporting_model}: tokens in/out = {tokens_in}/{tokens_out}\n")
+        if finish_reason != 'ok':
+            response += md(f"\n{finish_reason}\n")
         if tokens > 40000:
             response += "\n\nTHIS DIALOG IS GETTING TOO LONG. PLEASE RESTART CONVERSATION SOON."
         past.append({"role":"assistant", "content": final_text})
@@ -972,8 +980,7 @@ def show_help():
         4. "Speak Dialog" will voice whatever is currently in the Dialog window.  You can repeat it and you
              can edit what's to be spoken.  Except:  In a chat conversation, spoken dialog will only include
              the latest prompt/response ("YOU:/GPT:") sequence.'''
-    return md(txt)
-
+    return str(txt).replace('```', ' ').replace('  ', '&nbsp;&nbsp;').replace('  ', '&nbsp;&nbsp;').replace('  ', '&nbsp;&nbsp;').replace('\n','<br>')
 def upload_image(prompt, user, password):
     if not (user in unames and password == pwdList[unames.index(user)]):
         return [gr.Image(visible=False, interactive=True), "Incorrect user name and/or password"]
@@ -1165,9 +1172,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         speak_output = gr.Button(value="Speak Dialog", visible=True)
         submit_button = gr.Button(value="Submit Prompt/Question")
     prompt_window = gr.Textbox(label = "Prompt or Question")
-    #gr.Markdown('### **Dialog:**')
-    output_window = gr.Text(container=True, label='Dialog')
-    # output_window = gr.Markdown(container=True)
+    gr.Markdown('### **Dialog:**')
+    #output_window = gr.Text(container=True, label='Dialog')
+    output_window = gr.Markdown(container=True)
     with gr.Row():
         with gr.Column():
             image_window2 = gr.Image(visible=False, interactive=True, label='Image to Analyze', type='filepath')