Update LLM related Libs (#771)

backend/danswer/direct_qa/qa_block.py

@@ -220,7 +220,7 @@ class QABlock(QAModel):
             prompt_tokens = sum(
                 [
                     check_number_of_tokens(
-                        text=p.content, encode_fn=get_default_llm_token_encode()
+                        text=str(p.content), encode_fn=get_default_llm_token_encode()
                     )
                     for p in prompt
                 ]

backend/danswer/llm/chat_llm.py

@@ -43,7 +43,7 @@ class LangChainChatLLM(LLM, abc.ABC):
 
     def log_model_configs(self) -> None:
         logger.debug(
-            f"Model Class: {self.llm.__class__.__name__}, Model Config: {self.llm.__dict__}"
+            f"LLM Model Class: {self.llm.__class__.__name__}, Model Config: {self.llm.__dict__}"
         )
 
     def invoke(self, prompt: LanguageModelInput) -> str:
@@ -54,6 +54,12 @@ class LangChainChatLLM(LLM, abc.ABC):
         if LOG_ALL_MODEL_INTERACTIONS:
             logger.debug(f"Raw Model Output:\n{model_raw}")
 
+        if not isinstance(model_raw, str):
+            raise RuntimeError(
+                "Model output inconsistent with expected type, "
+                "is this related to a library upgrade?"
+            )
+
         return model_raw
 
     def stream(self, prompt: LanguageModelInput) -> Iterator[str]:

backend/danswer/llm/utils.py

@@ -137,6 +137,9 @@ def message_generator_to_string_generator(
     messages: Iterator[BaseMessageChunk],
 ) -> Iterator[str]:
     for message in messages:
+        if not isinstance(message.content, str):
+            raise RuntimeError("LLM message not in expected format.")
+
         yield message.content
 
 

backend/danswer/search/search_nlp_models.py

@@ -1,3 +1,4 @@
+import logging
 import os
 
 import numpy as np
@@ -30,6 +31,8 @@ from shared_models.model_server_models import RerankRequest
 from shared_models.model_server_models import RerankResponse
 
 logger = setup_logger()
+# Remove useless info about layer initialization
+logging.getLogger("transformers").setLevel(logging.ERROR)
 
 
 _TOKENIZER: None | AutoTokenizer = None

backend/requirements/default.txt

@@ -1,7 +1,7 @@
 alembic==1.10.4
 asyncpg==0.27.0
 atlassian-python-api==3.37.0
-beautifulsoup4==4.12.0
+beautifulsoup4==4.12.2
 celery==5.3.4
 dask==2023.8.1
 distributed==2023.8.1
@@ -21,13 +21,13 @@ httpx==0.23.3
 httpx-oauth==0.11.2
 huggingface-hub==0.16.4
 jira==3.5.1
-langchain==0.0.325
-litellm==0.12.5
-llama-index==0.8.27
+langchain==0.0.340
+litellm==1.7.5
+llama-index==0.9.8
 Mako==1.2.4
 nltk==3.8.1
 docx2txt==0.8
-openai==0.27.6
+openai==1.3.5
 oauthlib==3.2.2
 playwright==1.37.0
 psutil==5.9.5

backend/scripts/test-openapi-key.py

@@ -1,9 +1,22 @@
-from typing import cast
-
-import openai
+from openai import OpenAI
 
 
-VALID_MODEL_LIST = ["text-davinci-003", "gpt-3.5-turbo", "gpt-4"]
+VALID_MODEL_LIST = [
+    "gpt-4-1106-preview",
+    "gpt-4-vision-preview",
+    "gpt-4",
+    "gpt-4-0314",
+    "gpt-4-0613",
+    "gpt-4-32k",
+    "gpt-4-32k-0314",
+    "gpt-4-32k-0613",
+    "gpt-3.5-turbo-1106",
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-0301",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-16k-0613",
+]
 
 
 if __name__ == "__main__":
@@ -12,29 +25,28 @@ if __name__ == "__main__":
         model_version = input("Please provide an OpenAI model version to test: ")
         if model_version not in VALID_MODEL_LIST:
             print(f"Model must be from valid list: {', '.join(VALID_MODEL_LIST)}")
+    assert model_version
 
     api_key = input("Please provide an OpenAI API Key to test: ")
-    openai.api_key = api_key
+    client = OpenAI(
+        api_key=api_key,
+    )
 
     prompt = "The boy went to the "
     print(f"Asking OpenAI to finish the sentence using {model_version}")
     print(prompt)
     try:
-        if model_version == "text-davinci-003":
-            response = openai.Completion.create(
-                model=model_version, prompt=prompt, max_tokens=5, temperature=2
-            )
-            print(cast(str, response["choices"][0]["text"]).strip())
-
-        else:
-            messages = [
-                {"role": "system", "content": "Finish the sentence"},
-                {"role": "user", "content": prompt},
-            ]
-            response = openai.ChatCompletion.create(
-                model=model_version, messages=messages, max_tokens=5, temperature=2
-            )
-            print(cast(str, response["choices"][0]["message"]["content"]).strip())
+        messages = [
+            {"role": "system", "content": "Finish the sentence"},
+            {"role": "user", "content": prompt},
+        ]
+        response = client.chat.completions.create(
+            model=model_version,
+            messages=messages,  # type:ignore
+            max_tokens=5,
+            temperature=2,
+        )
+        print(response.choices[0].message.content)
         print("Success! Feel free to use this API key for Danswer.")
     except Exception:
         print(