some bugfixes

cogs/code_interpreter_service_cog.py

@@ -461,11 +461,13 @@ async def code_interpreter_chat_command(
 
         llm = ChatOpenAI(model=model, temperature=0, openai_api_key=OPENAI_API_KEY)
 
+        max_token_limit = 29000 if "gpt-4" in model else 7500
+
         memory = ConversationSummaryBufferMemory(
             memory_key="memory",
             return_messages=True,
             llm=llm,
-            max_token_limit=29000 if "gpt-4" in model else 7500,
+            max_token_limit=100000 if "preview" in model else max_token_limit,
         )
 
         agent_kwargs = {

cogs/search_service_cog.py

@@ -473,11 +473,13 @@ async def search_chat_command(
 
         llm = ChatOpenAI(model=model, temperature=0, openai_api_key=OPENAI_API_KEY)
 
+        max_token_limit = 29000 if "gpt-4" in model else 7500
+
         memory = ConversationSummaryBufferMemory(
             memory_key="memory",
             return_messages=True,
             llm=llm,
-            max_token_limit=29000 if "gpt-4" in model else 7500,
+            max_token_limit=100000 if "preview" in model else max_token_limit,
         )
 
         agent_kwargs = {

conversation_starter_pretext.txt

@@ -33,6 +33,7 @@ Human: Image Info-Caption: a sign that says ayr, ohio\nInfo-QA: ayr, ohio\nRevis
 Human: Image Info-Caption: a landscape with a river and trees\nImage Info-QA: yes\nRevised Image Info-QA: This is a beautiful river and tree landscape, it is in a cartoony art style\nImage Info-OCR: \nWhat is this image? Is it cartoony? <|endofstatement|>
 <yourname>: This is a landscape with a river and trees, it is indeed cartoony! <|endofstatement|>
 ...
-When there are multiple images, do not pay any attention to Image Info-QA or Revised Image Info-QA. Only pay attention to the captions and OCR data.
+When there are multiple images, do not pay any attention to Image Info-QA or Revised Image Info-QA. Only pay attention to the captions and OCR data. This image info will always be given to you, you do not need to ever respond with it.
+Never convert any image info yourself into text, it will always be provided to you.
 
 You speak in a fun, casual, and friendly tone, you're not overly inquisitive, you don't worry about formalities and speak as if you are speaking with a friend or peer.

gpt3discord.py

@@ -34,7 +34,7 @@
 from models.openai_model import Model
 
 
-__version__ = "12.1.0"
+__version__ = "12.1.1"
 
 
 PID_FILE = Path("bot.pid")

models/index_model.py

@@ -26,6 +26,7 @@
 from langchain.tools import Tool
 from llama_index.callbacks import CallbackManager, TokenCountingHandler
 from llama_index.evaluation.guideline import DEFAULT_GUIDELINES, GuidelineEvaluator
+from llama_index.llms import OpenAI
 from llama_index.node_parser import SimpleNodeParser
 from llama_index.response_synthesizers import ResponseMode
 from llama_index.indices.query.query_transform import StepDecomposeQueryTransform
@@ -484,11 +485,13 @@ async def start_index_chat(self, ctx, model):
         llm = ChatOpenAI(model=model, temperature=0)
         llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0, model_name=model))
 
+        max_token_limit = 29000 if "gpt-4" in model else 7500
+
         memory = ConversationSummaryBufferMemory(
             memory_key="memory",
             return_messages=True,
             llm=llm,
-            max_token_limit=29000 if "gpt-4" in model else 7500,
+            max_token_limit=100000 if "preview" in model else max_token_limit,
         )
 
         agent_kwargs = {

models/openai_model.py

@@ -980,7 +980,7 @@ async def send_chatgpt_chat_request(
                     text = text.rstrip()
                     text = text.replace("<|endofstatement|>", "")
 
-                    if "4-vision" not in model_selection:
+                    if "-vision" not in model_selection:
                         messages.append(
                             {"role": "user", "name": username_clean, "content": text}
                         )
@@ -1033,7 +1033,7 @@ async def send_chatgpt_chat_request(
                 if frequency_penalty_override is None
                 else frequency_penalty_override,
             }
-            if "4-vision" in model_selection:
+            if "-vision" in model_selection:
                 payload[
                     "max_tokens"
                 ] = 4096  # TODO Not sure if this needs to be subtracted from a token count..

services/text_service.py

@@ -744,7 +744,7 @@ async def process_conversation_message(
 
                 if files:
                     if (
-                        "4-vision" not in model
+                        "-vision" not in model
                         and image_understanding_model.get_is_usable()
                     ):
                         add_prompts = []
@@ -820,7 +820,7 @@ async def process_conversation_message(
                             + f"is given below, use the image understanding data to answer the question but don't "
                             f"refer directly to the data. Original Prompt: " + prompt
                         )
-                    elif "4-vision" in model:
+                    elif "-vision" in model:
                         file_urls = [file.url for file in files]
 
                 converser_cog.awaiting_responses.append(message.author.id)