use 4o-mini in more places

app/pipeline/chat/course_chat_pipeline.py

@@ -70,7 +70,8 @@ def get_mastery(progress, confidence):
 class CourseChatPipeline(Pipeline):
     """Course chat pipeline that answers course related questions from students."""
 
-    llm: IrisLangchainChatModel
+    llm_big: IrisLangchainChatModel
+    llm_small: IrisLangchainChatModel
     pipeline: Runnable
     lecture_pipeline: LectureChatPipeline
     suggestion_pipeline: InteractionSuggestionPipeline
@@ -93,14 +94,20 @@ def __init__(
         self.event = event
 
         # Set the langchain chat model
-        request_handler = CapabilityRequestHandler(
-            requirements=RequirementList(
-                gpt_version_equivalent=4.5,
-            )
-        )
         completion_args = CompletionArguments(temperature=0.5, max_tokens=2000)
-        self.llm = IrisLangchainChatModel(
-            request_handler=request_handler, completion_args=completion_args
+        self.llm_big = IrisLangchainChatModel(
+            request_handler=CapabilityRequestHandler(
+                requirements=RequirementList(
+                    gpt_version_equivalent=4.5,
+                )
+            ), completion_args=completion_args
+        )
+        self.llm_small = IrisLangchainChatModel(
+            request_handler=CapabilityRequestHandler(
+                requirements=RequirementList(
+                    gpt_version_equivalent=4.25,
+                )
+            ), completion_args=completion_args
         )
         self.callback = callback
 
@@ -110,14 +117,14 @@ def __init__(
         self.citation_pipeline = CitationPipeline()
 
         # Create the pipeline
-        self.pipeline = self.llm | JsonOutputParser()
+        self.pipeline = self.llm_big | JsonOutputParser()
         self.tokens = []
 
     def __repr__(self):
-        return f"{self.__class__.__name__}(llm={self.llm})"
+        return f"{self.__class__.__name__}(llm_big={self.llm_big}, llm_small={self.llm_small})"
 
     def __str__(self):
-        return f"{self.__class__.__name__}(llm={self.llm})"
+        return f"{self.__class__.__name__}(llm_big={self.llm_big}, llm_small={self.llm_small})"
 
     @traceable(name="Course Chat Pipeline")
     def __call__(self, dto: CourseChatPipelineExecutionDTO, **kwargs):
@@ -395,7 +402,7 @@ def lecture_content_retrieval() -> str:
             # No idea why we need this extra contrary to exercise chat agent in this case, but solves the issue.
             params.update({"tools": tools})
             agent = create_tool_calling_agent(
-                llm=self.llm, tools=tools, prompt=self.prompt
+                llm=self.llm_big, tools=tools, prompt=self.prompt
             )
             agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=False)
 
@@ -404,7 +411,7 @@ def lecture_content_retrieval() -> str:
             for step in agent_executor.iter(params):
                 print("STEP:", step)
                 self._append_tokens(
-                    self.llm.tokens, PipelineEnum.IRIS_CHAT_COURSE_MESSAGE
+                    self.llm_big.tokens, PipelineEnum.IRIS_CHAT_COURSE_MESSAGE
                 )
                 if step.get("output", None):
                     out = step["output"]

app/pipeline/chat/exercise_chat_agent_pipeline.py

@@ -94,7 +94,8 @@ def map_message_role(role: IrisMessageRole) -> str:
 class ExerciseChatAgentPipeline(Pipeline):
     """Exercise chat agent pipeline that answers exercises related questions from students."""
 
-    llm: IrisLangchainChatModel
+    llm_big: IrisLangchainChatModel
+    llm_small: IrisLangchainChatModel
     pipeline: Runnable
     callback: ExerciseChatStatusCallback
     suggestion_pipeline: InteractionSuggestionPipeline
@@ -112,14 +113,22 @@ def __init__(
         super().__init__(implementation_id="exercise_chat_pipeline")
         # Set the langchain chat model
         completion_args = CompletionArguments(temperature=0.5, max_tokens=2000)
-        self.llm = IrisLangchainChatModel(
+        self.llm_big = IrisLangchainChatModel(
             request_handler=CapabilityRequestHandler(
                 requirements=RequirementList(
                     gpt_version_equivalent=4.5,
                 ),
             ),
             completion_args=completion_args,
         )
+        self.llm_small = IrisLangchainChatModel(
+            request_handler=CapabilityRequestHandler(
+                requirements=RequirementList(
+                    gpt_version_equivalent=4.25,
+                ),
+            ),
+            completion_args=completion_args,
+        )
         self.variant = variant
         self.event = event
         self.callback = callback
@@ -130,15 +139,15 @@ def __init__(
         self.retriever = LectureRetrieval(self.db.client)
         self.reranker_pipeline = RerankerPipeline()
         self.code_feedback_pipeline = CodeFeedbackPipeline()
-        self.pipeline = self.llm | JsonOutputParser()
+        self.pipeline = self.llm_big | JsonOutputParser()
         self.citation_pipeline = CitationPipeline()
         self.tokens = []
 
     def __repr__(self):
-        return f"{self.__class__.__name__}(llm={self.llm})"
+        return f"{self.__class__.__name__}(llm_big={self.llm_big}, llm_small={self.llm_small})"
 
     def __str__(self):
-        return f"{self.__class__.__name__}(llm={self.llm})"
+        return f"{self.__class__.__name__}(llm_big={self.llm_big}, llm_small={self.llm_small})"
 
     @traceable(name="Exercise Chat Agent Pipeline")
     def __call__(self, dto: ExerciseChatPipelineExecutionDTO):
@@ -504,15 +513,14 @@ def lecture_content_retrieval() -> str:
                 tool_list.append(lecture_content_retrieval)
             tools = generate_structured_tools_from_functions(tool_list)
             agent = create_tool_calling_agent(
-                llm=self.llm, tools=tools, prompt=self.prompt
+                llm=self.llm_big, tools=tools, prompt=self.prompt
             )
             agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=False)
             self.callback.in_progress()
             out = None
             for step in agent_executor.iter(params):
-                print("STEP:", step)
                 self._append_tokens(
-                    self.llm.tokens, PipelineEnum.IRIS_CHAT_EXERCISE_AGENT_MESSAGE
+                    self.llm_big.tokens, PipelineEnum.IRIS_CHAT_EXERCISE_AGENT_MESSAGE
                 )
                 if step.get("output", None):
                     out = step["output"]
@@ -525,13 +533,13 @@ def lecture_content_retrieval() -> str:
                     ]
                 )
 
-                guide_response = (self.prompt | self.llm | StrOutputParser()).invoke(
+                guide_response = (self.prompt | self.llm_small | StrOutputParser()).invoke(
                     {
                         "response": out,
                     }
                 )
                 self._append_tokens(
-                    self.llm.tokens, PipelineEnum.IRIS_CHAT_EXERCISE_AGENT_MESSAGE
+                    self.llm_small.tokens, PipelineEnum.IRIS_CHAT_EXERCISE_AGENT_MESSAGE
                 )
                 if "!ok!" in guide_response:
                     print("Response is ok and not rewritten!!!")

app/pipeline/chat/interaction_suggestion_pipeline.py

@@ -61,7 +61,7 @@ def __init__(self, variant: str = "default"):
         # Set the langchain chat model
         request_handler = CapabilityRequestHandler(
             requirements=RequirementList(
-                gpt_version_equivalent=4.5,
+                gpt_version_equivalent=4.25,
                 json_mode=True,
             )
         )

app/pipeline/prompts/citation_prompt.txt

@@ -28,5 +28,5 @@ Answer without citations:
 Paragraphs with their Lecture Names, Unit Names, Links and Page Numbers:
 {Paragraphs}
 
-Answer with citations (ensure empty line between the message and the citations):
 If the answer actually does not contain any information from the paragraphs, please do not include any citations and return '!NONE!'.
+Original answer with citations (ensure two empty lines between the message and the citations):