Update app.py

4 deepseek models via together.ai

app.py

@@ -32,6 +32,7 @@ users = os.getenv('LOGNAME')
 unames = users.split(',')
 pwds = os.getenv('PASSWORD')
 pwdList = pwds.split(',')
+DEEPSEEK_KEY=os.getenv('DEEPSEEK_KEY')
 
 site = os.getenv('SITE')
 if site == 'local':
@@ -660,8 +661,25 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
             return [past, 'Stock data file created', None, gptModel, uploaded_image_file, plot]
     if user_window in unames and pwd_window == pwdList[unames.index(user_window)]:
         chatType = 'normal'
+        deepseek = False
         prompt = prompt.strip()
-        if prompt.lower().startswith('solve'):
+        if prompt.lower().startswith('dsr1 '):
+            deepseek = True
+            ds_model = 'deepseek-ai/DeepSeek-R1'
+            prompt = prompt[5:]
+        elif prompt.lower().startswith('ds1.5 '):
+            deepseek = True
+            ds_model = 'deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B'
+            prompt = prompt[6:]
+        elif prompt.lower().startswith('ds14 '):
+            deepseek = True
+            ds_model = 'deepseek-ai/DeepSeek-R1-Distill-Qwen-14B'
+            prompt = prompt[5:]
+        elif prompt.lower().startswith('ds70 '):
+            deepseek = True
+            ds_model = 'deepseek-ai/DeepSeek-R1-Distill-Llama-70B'
+            prompt = prompt[5:]
+        elif prompt.lower().startswith('solve'):
             prompt = 'How do I solve ' + prompt[5:] + ' Do not use Latex for math expressions.'
             chatType = 'math'
         elif prompt.lower().startswith('puzzle'):
@@ -673,9 +691,16 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
            (reply, tokens_in, tokens_out, tokens) = solve(prompt, chatType)
            reporting_model = image_gen_model
         elif not gen_image:
-            completion = Client().chat.completions.create(model=gptModel,
+            if deepseek:
+                client = OpenAI(api_key=DEEPSEEK_KEY, base_url='https://api.together.xyz/v1')
+                completion = client.chat.completions.create(
+                    model= ds_model,
+                    messages=past)
+                reporting_model='deepseek'
+            else:
+                completion = Client().chat.completions.create(model=gptModel,
                                             messages=past)
-            reporting_model = gptModel
+                reporting_model = gptModel
         else:
             (completion, msg) = analyze_image(user_window, image_gen_model, prompt)
             uploaded_image_file= ''
@@ -684,6 +709,14 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
                 return [past, msg, None, gptModel, uploaded_image_file, plot]
         if not chatType in special_chat_types:
             reply = completion.choices[0].message.content
+            final_text = reply
+            if deepseek:
+                loc1 = reply.find('<think>')
+                if loc1 > -1:
+                    loc2 = reply.find('</think>')
+                    if loc2 > loc1:
+                        final_text = reply[loc2 + 8:]
+                reply = reply.replace('<think>','Thinking:\n').replace('</think>','Done thinking:\n')
             tokens_in = completion.usage.prompt_tokens
             tokens_out = completion.usage.completion_tokens
             tokens = completion.usage.total_tokens
@@ -692,22 +725,23 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
             response += md(f"\n{reporting_model}: tokens in/out = {tokens_in}/{tokens_out}")
         if tokens > 40000:
             response += "\n\nTHIS DIALOG IS GETTING TOO LONG. PLEASE RESTART CONVERSATION SOON."
-        past.append({"role":"assistant", "content": reply})
-        accessOk = False
-        for i in range(3):
-            try:
-                dataFile = new_func(user_window)
-                with open(dataFile, 'a') as f:
-                    m = '4o'
-                    if 'mini' in reporting_model:
-                        m = '4omini'
-                    f.write(f'{user_window}:{tokens_in}/{tokens_out}-{m}\n')
-                accessOk = True
-                break
-            except Exception as e:
-                sleep(3)
-        if not accessOk:
-            response += f"\nDATA LOG FAILED, path = {dataFile}"
+        past.append({"role":"assistant", "content": final_text})
+        if not deepseek:
+            accessOk = False
+            for i in range(3):
+                try:
+                    dataFile = new_func(user_window)
+                    with open(dataFile, 'a') as f:
+                        m = '4o'
+                        if 'mini' in reporting_model:
+                            m = '4omini'
+                        f.write(f'{user_window}:{tokens_in}/{tokens_out}-{m}\n')
+                    accessOk = True
+                    break
+                except Exception as e:
+                    sleep(3)
+            if not accessOk:
+                response += f"\nDATA LOG FAILED, path = {dataFile}"
         return [past, response , None, gptModel, uploaded_image_file, plot]
     else:
         return [[], "User name and/or password are incorrect", prompt, gptModel, uploaded_image_file, plot]