From 38a8f8f51894436f2dce1f0dc051a0cc0e632078 Mon Sep 17 00:00:00 2001
From: NotBioWaste905 <andruhabobr@gmail.com>
Date: Tue, 30 Jul 2024 18:49:28 +0300
Subject: [PATCH] Figured out how to implement DeepEval functions

---
 chatsky/llm/conditions.py                   | 14 ++++++++++++++
 chatsky/llm/{llm_response.py => wrapper.py} | 19 +++++++------------
 2 files changed, 21 insertions(+), 12 deletions(-)
 create mode 100644 chatsky/llm/conditions.py
 rename chatsky/llm/{llm_response.py => wrapper.py} (89%)

diff --git a/chatsky/llm/conditions.py b/chatsky/llm/conditions.py
new file mode 100644
index 000000000..c393d7a6c
--- /dev/null
+++ b/chatsky/llm/conditions.py
@@ -0,0 +1,14 @@
+"""
+LLM conditions.
+---------
+In this file stored unified functions for some basic condition cases
+including regex search, semantic distance (cosine) etc.
+"""
+
+from chatsky.script.core.message import Message
+
+def regex_search(pattern: str) -> bool:
+    pass
+
+def semantic_distance(target: str | Message, threshold: float) -> bool:
+    pass
\ No newline at end of file
diff --git a/chatsky/llm/llm_response.py b/chatsky/llm/wrapper.py
similarity index 89%
rename from chatsky/llm/llm_response.py
rename to chatsky/llm/wrapper.py
index 6aa2f8127..8e1c12a68 100644
--- a/chatsky/llm/llm_response.py
+++ b/chatsky/llm/wrapper.py
@@ -11,13 +11,12 @@
     from langchain_cohere import ChatCohere
     from langchain_mistralai import ChatMistralAI
     from langchain.output_parsers import ResponseSchema, StructuredOutputParser
+    from langchain_core.messages import HumanMessage, SystemMessage
+    from langchain_core.output_parsers import StrOutputParser
     langchain_available = True
 except ImportError:
     langchain_available = False
 
-from langchain_core.messages import HumanMessage, SystemMessage
-from langchain_core.output_parsers import StrOutputParser
-
 import base64
 import httpx
 import re
@@ -78,15 +77,11 @@ def condition(self, prompt, request):
         return result
     
     # Helper functions for DeepEval custom LLM usage
-    def generate(self, prompt: str, schema: BaseModel):
-        # TODO: Remake this
-        schema_parser = StructuredOutputParser.from_response_schemas([ResponseSchema(base_model=schema)])
-        chain = prompt | self.model | schema_parser
-        return chain.invoke({"input": prompt})
+    def generate(self, prompt: str):
+        return self.model.invoke(prompt).content
     
-    async def a_generate(self, prompt: str, schema: BaseModel):
-        # TODO: Remake this
-        return self.generate(HumanMessage(prompt), schema)
+    async def a_generate(self, prompt: str):
+        return self.generate(prompt)
     
     def load_model(self):
         return self.model
@@ -174,4 +169,4 @@ def message_to_langchain(message: Message):
             content.append(
                 {"type": "image_url", "image_url": {"url": __attachment_to_content(image)}}
             )
-    return HumanMessage(content=content)
\ No newline at end of file
+    return HumanMessage(content=content)