Together AI issues (#185)

* Ensure all configurations work.

app/transcribe/args.py

@@ -85,18 +85,22 @@ def handle_args_batch_tasks(args: argparse.Namespace, global_vars: Transcription
     if args.validate_api_key is not None:
         chat_inference_provider = config['General']['chat_inference_provider']
         if chat_inference_provider == 'openai':
-            base_url = config['OpenAI']['base_url']
+            settings_section = 'OpenAI'
         elif chat_inference_provider == 'together':
-            base_url = config['Together']['base_url']
+            settings_section = 'Together'
 
-        if utilities.is_api_key_valid(api_key=args.validate_api_key, base_url=base_url):
+        base_url = config[settings_section]['base_url']
+        model = config[settings_section]['ai_model']
+
+        if utilities.is_api_key_valid(api_key=args.validate_api_key, base_url=base_url, model=model):
             print('The api_key is valid')
-            base_url = config['OpenAI']['base_url']
             client = openai.OpenAI(api_key=args.validate_api_key, base_url=base_url)
-            models = utilities.get_available_models(client=client)
-            print('Available models: ')
-            for model in models:
-                print(f'    {model}')
+
+            if base_url != 'https://api.together.xyz':
+                models = utilities.get_available_models(client=client)
+                print('Available models: ')
+                for model in models:
+                    print(f'    {model}')
             client.close()
         else:
             print('The api_key is not valid')

app/transcribe/gpt_responder.py

@@ -7,7 +7,7 @@
 import conversation
 import constants
 from tsutils import app_logging as al
-from tsutils import duration
+from tsutils import duration, utilities
 
 
 root_logger = al.get_logger()
@@ -47,8 +47,17 @@ def summarize(self) -> str:
         """
         root_logger.info(GPTResponder.summarize.__name__)
 
-        if self.config['OpenAI']['api_key'] in ('', 'API_KEY'):
-            # Cannot summarize without connection to LLM
+        chat_inference_provider = self.config['General']['chat_inference_provider']
+        if chat_inference_provider == 'openai':
+            settings_section = 'OpenAI'
+        elif chat_inference_provider == 'together':
+            settings_section = 'Together'
+
+        api_key = self.config[settings_section]['api_key']
+        base_url = self.config[settings_section]['base_url']
+        model = self.config[settings_section]['ai_model']
+
+        if not utilities.is_api_key_valid(api_key=api_key, base_url=base_url, model=model):
             return None
 
         with duration.Duration(name='OpenAI Summarize', screen=False):
@@ -81,7 +90,18 @@ def generate_response_from_transcript_no_check(self) -> str:
         """
         try:
             root_logger.info(GPTResponder.generate_response_from_transcript_no_check.__name__)
-            if self.config['OpenAI']['api_key'] in ('', 'API_KEY'):
+            chat_inference_provider = self.config['General']['chat_inference_provider']
+
+            if chat_inference_provider == 'openai':
+                settings_section = 'OpenAI'
+            elif chat_inference_provider == 'together':
+                settings_section = 'Together'
+
+            api_key = self.config[settings_section]['api_key']
+            base_url = self.config[settings_section]['base_url']
+            model = self.config[settings_section]['ai_model']
+
+            if not utilities.is_api_key_valid(api_key=api_key, base_url=base_url, model=model):
                 return None
 
             with duration.Duration(name='OpenAI Chat Completion', screen=False):
@@ -172,7 +192,17 @@ def generate_response_for_selected_text(self, text: str):
         """
         try:
             root_logger.info(GPTResponder.generate_response_for_selected_text.__name__)
-            if self.config['OpenAI']['api_key'] in ('', 'API_KEY'):
+            chat_inference_provider = self.config['General']['chat_inference_provider']
+            if chat_inference_provider == 'openai':
+                settings_section = 'OpenAI'
+            elif chat_inference_provider == 'together':
+                settings_section = 'Together'
+
+            api_key = self.config[settings_section]['api_key']
+            base_url = self.config[settings_section]['base_url']
+            model = self.config[settings_section]['ai_model']
+
+            if not utilities.is_api_key_valid(api_key=api_key, base_url=base_url, model=model):
                 return None
 
             with duration.Duration(name='OpenAI Chat Completion Selected', screen=False):
@@ -305,7 +335,7 @@ def __init__(self,
         self.llm_client = openai.OpenAI(api_key=api_key,
                                         base_url=base_url)
         self.model = self.config['Together']['ai_model']
-        print(f'[INFO] Using Together for inference. Model: {self.model}')
+        print(f'[INFO] Using Together AI for inference. Model: {self.model}')
         super().__init__(config=self.config,
                          convo=convo,
                          save_to_file=save_to_file,

app/transcribe/ui.py

@@ -487,11 +487,13 @@ def create_ui_components(root, config: dict):
     if chat_inference_provider == 'openai':
         api_key = config['OpenAI']['api_key']
         base_url = config['OpenAI']['base_url']
+        model = config['OpenAI']['ai_model']
     elif chat_inference_provider == 'together':
         api_key = config['Together']['api_key']
         base_url = config['Together']['base_url']
+        model = config['Together']['ai_model']
 
-    if not utilities.is_api_key_valid(api_key=api_key, base_url=base_url):
+    if not utilities.is_api_key_valid(api_key=api_key, base_url=base_url, model=model):
         # Disable buttons that interact with backend services
         continuous_response_button.configure(state='disabled')
         response_now_button.configure(state='disabled')

examples/together/stt.py

@@ -15,11 +15,12 @@
     },
     {
       'role': 'user',
-      'content': 'Tell me about Fantasy Football',
+      'content': 'Are you online',
     }
   ],
   model='mistralai/Mixtral-8x7B-Instruct-v0.1',
   max_tokens=1024
 )
 
 print(chat_completion.choices[0].message.content)
+client.close()

tsutils/utilities.py

@@ -6,6 +6,9 @@
 import openai
 
 
+valid_api_key: bool = False
+
+
 def merge(first: dict, second: dict, path=[]):
     """Recursively merge two dictionaries.
        For keys with different values, values in the second dictionary
@@ -207,18 +210,41 @@ def get_available_models(client: openai.OpenAI) -> list:
     return sorted(return_val)
 
 
-def is_api_key_valid(api_key: str, base_url) -> bool:
+def is_api_key_valid(api_key: str, base_url: str, model: str) -> bool:
     """Check if it is valid openai compatible openai key for the provider
     """
-    openai.api_key = api_key
 
+    global valid_api_key  # pylint: disable=W0603
+
+    if valid_api_key:
+        return True
+
+    openai.api_key = api_key
     client = openai.OpenAI(api_key=api_key, base_url=base_url)
 
     try:
-        client.models.list()
+        # Ideally models list is the best way to determine if api key is valid
+        # Some of the OpenAI compatible vendors do not support all the methods though
+        # client.models.list()
+        chat_completion = client.chat.completions.create(
+            messages=[
+                {
+                    'role': 'system',
+                    'content': 'You are an AI assistant',
+                },
+                {
+                    'role': 'user',
+                    'content': 'Are you online',
+                }
+            ],
+            model=model,
+            max_tokens=1024
+            )
+        assert len(chat_completion.choices[0].message.content) > 0
         client.close()
     except openai.AuthenticationError as e:
         print(e)
         return False
 
+    valid_api_key = True
     return True