Merge pull request #32 from ijwfly/feature/gpt-4o

GPT-4o Support

README.md

@@ -2,7 +2,7 @@
 
 This GitHub repository contains the implementation of a telegram bot, designed to facilitate seamless interaction with GPT-3.5 and GPT-4, state-of-the-art language models by OpenAI.  
 
-🔥 **GPT-4 Turbo support (with vision)**  
+🔥 **GPT-4o support (with vision)**  
 🔥 **Custom OpenAI API compatible endpoints support (see `app/llm_models.py` for example of using WizardLM-2 8x22b via OpenRouter.ai)**  
 🔥 **DALL-E 3 Image generation support**
 

app/bot/settings_menu.py

@@ -7,8 +7,9 @@
 from app.storage.user_role import check_access_conditions, UserRole
 
 GPT_MODELS_OPTIONS = {
-    'gpt-3.5-turbo': 'GPT-3.5',
+    'gpt-4o': 'GPT-4o',
     'gpt-4-turbo': 'GPT-4-Turbo',
+    'gpt-3.5-turbo': 'GPT-3.5',
 }
 
 ALL_MODELS_OPTIONS = list(get_models().keys())

app/llm_models.py

@@ -37,6 +37,7 @@ class LLModel:
     GPT_35_TURBO_16K = 'gpt-3.5-turbo-16k'
     GPT_4 = 'gpt-4'
     GPT_4_TURBO = 'gpt-4-turbo'
+    GPT_4O = 'gpt-4o'
     GPT_4_TURBO_PREVIEW = 'gpt-4-turbo-preview'
     GPT_4_VISION_PREVIEW = 'gpt-4-vision-preview'
     OPENROUTER_WIZARDLM2 = 'microsoft/wizardlm-2-8x22b'
@@ -109,6 +110,27 @@ def get_models():
                 ),
                 base_url=settings.OPENAI_BASE_URL,
             ),
+            LLModel.GPT_4O: LLModel(
+                model_name=LLModel.GPT_4O,
+                model_readable_name='GPT-4o',
+                api_key=settings.OPENAI_TOKEN,
+                minimum_user_role=settings.USER_ROLE_CHOOSE_MODEL,
+                context_configuration=LLMContextConfiguration(
+                    short_term_memory_tokens=8 * 1024,
+                    summary_length=2048,
+                    hard_max_context_size=13 * 1024,
+                ),
+                model_price=LLMPrice(
+                    input_tokens_price=Decimal('0.005'),
+                    output_tokens_price=Decimal('0.015'),
+                ),
+                capabilities=LLMCapabilities(
+                    function_calling=True,
+                    image_processing=True,
+                    streaming_responses=True,
+                ),
+                base_url=settings.OPENAI_BASE_URL,
+            ),
 
             # Deprecated OpenAI models
             LLModel.GPT_35_TURBO_16K: LLModel(

app/openai_helpers/count_tokens.py

@@ -20,6 +20,8 @@ def count_string_tokens(string: str, model="gpt-3.5-turbo") -> int:
         model = "gpt-3.5-turbo"
     elif "gpt-4" in model:
         model = "gpt-4"
+    elif "gpt-4o" in model:
+        model = "gpt-4o"
     else:
         # TODO: add method to calculate tokens for different models
         model = "gpt-4"

requirements.txt

@@ -27,7 +27,7 @@ pytz==2023.3
 regex==2023.6.3
 requests==2.31.0
 sniffio==1.3.0
-tiktoken==0.4.0
+tiktoken==0.7.0
 tqdm==4.65.0
 typing_extensions==4.8.0
 urllib3==2.0.3