feat(openai): 16k

backend/llm/qa.py

@@ -27,7 +27,7 @@ class CustomSupabaseVectorStore(SupabaseVectorStore):
         query: str, 
         user_id: str = "none",
         table: str = "match_vectors", 
-        k: int = 4, 
+        k: int = 8, 
         threshold: float = 0.5, 
         **kwargs: Any
     ) -> List[Document]:

backend/llm/summarization.py

@@ -2,7 +2,6 @@ import os
 
 import guidance
 import openai
-
 from logger import get_logger
 
 logger = get_logger(__name__)

backend/models/chats.py

@@ -1,11 +1,11 @@
-from typing import List, Tuple, Optional
+from typing import List, Optional, Tuple
 from uuid import UUID
 
 from pydantic import BaseModel
 
 
 class ChatMessage(BaseModel):
-    model: str = "gpt-3.5-turbo-0613"
+    model: str = "gpt-3.5-turbo-16k"
     question: str
     # A list of tuples where each tuple is (speaker, text)
     history: List[Tuple[str, str]]

backend/parsers/audio.py

@@ -58,7 +58,7 @@ async def process_audio(upload_file: UploadFile, enable_summarization: bool, use
     file_size = len(transcript.text.encode("utf-8"))
 
     # Load chunk size and overlap from sidebar
-    chunk_size = 500
+    chunk_size = 250
     chunk_overlap = 0
 
     text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(

backend/parsers/common.py

@@ -31,7 +31,7 @@ async def process_file(file: UploadFile, loader_class, file_suffix, enable_summa
         file_sha1 = compute_sha1_from_file(tmp_file.name)
 
     os.remove(tmp_file.name)
-    chunk_size = 500
+    chunk_size = 250
     chunk_overlap = 0
 
     text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(

backend/parsers/github.py

@@ -22,7 +22,7 @@ async def process_github(repo, enable_summarization, user, supabase, user_openai
     documents = loader.load()
     os.system("rm -rf /tmp/" + random_dir_name)
 
-    chunk_size = 500
+    chunk_size = 250
     chunk_overlap = 0
     text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(
         chunk_size=chunk_size, chunk_overlap=chunk_overlap)

frontend/lib/context/BrainConfigProvider/brain-config-provider.tsx

@@ -16,9 +16,9 @@ export const BrainConfigContext = createContext<ConfigContext | undefined>(
 );
 
 const defaultBrainConfig: BrainConfig = {
-  model: "gpt-3.5-turbo-0613",
+  model: "gpt-3.5-turbo-16k",
   temperature: 0,
-  maxTokens: 500,
+  maxTokens: 1000,
   keepLocal: true,
   anthropicKey: undefined,
   backendUrl: undefined,

frontend/lib/context/BrainConfigProvider/types.ts

@@ -21,14 +21,14 @@ export type ConfigContext = {
 // export const openAiModels = ["gpt-3.5-turbo", "gpt-4"] as const; ## TODO activate GPT4 when not in demo mode
 
 export const openAiModels = [
+  "gpt-3.5-turbo-16k",
   "gpt-3.5-turbo",
   "gpt-3.5-turbo-0613",
-  "gpt-3.5-turbo-16k",
 ] as const;
 export const openAiPaidModels = [
+  "gpt-3.5-turbo-16k",
   "gpt-3.5-turbo",
   "gpt-3.5-turbo-0613",
-  "gpt-3.5-turbo-16k",
   "gpt-4",
   "gpt-4-0613",
 ] as const;