fixes

chains/learning_objectives_generator/learning_objectives_chain.py

@@ -6,76 +6,7 @@ from langchain_core.prompts.chat import ChatPromptTemplate
 
 class LearningObjectivesChain(BaseModel):
     """
-
     """
 
-    # Templates
-    template_generator: ChatPromptTemplate
-    template_eliminator: ChatPromptTemplate
-    template_finetuner: ChatPromptTemplate
-    template_presenter: ChatPromptTemplate
-
-    # LLM references
-    llm_main: Any   # The "Main LLM"
-    llm_alt: Any    # The "Other LLM" used only for the second generator prompt
-
-    async def run(self, standardized_studytext: str) -> str:
-        """
-        Main pipeline for a single run. The 'standardized_studytext' is assumed
-        to be already standardized outside of this chain, so we jump straight to generation.
-        """
-
-        # 1) Two parallel calls to learning_objective_generator
-        #    with different LLMs (llm_main vs. llm_alt).
-
-        async def run_generator(llm, llm_label) -> str:
-            # Format the generator prompt
-            prompt = await self.template_generator.aformat_prompt(
-                standardized_studytext=standardized_studytext,  # possibly other variables
-                llm_label=llm_label
-            )
-            messages = prompt.to_messages()
-            response = await llm.ainvoke(messages)
-            return getattr(response, "content", response)
-
-        # Launch in parallel
-        gen_main_task = asyncio.create_task(run_generator(self.llm_main, "MAIN"))
-        gen_alt_task  = asyncio.create_task(run_generator(self.llm_alt, "ALT"))
-
-        # Wait for both to finish
-        gen_main_result, gen_alt_result = await asyncio.gather(gen_main_task, gen_alt_task)
-
-        # Combine them
-        combined_generators = f"[GEN MAIN]\n{gen_main_result}\n\n[GEN ALT]\n{gen_alt_result}"
-
-        # 2) learning_objective_eliminator (llm_main)
-        prompt_eliminate = await self.template_eliminator.aformat_prompt(
-            combined_generators=combined_generators,
-            standardized_studytext=standardized_studytext
-        )
-        elim_messages = prompt_eliminate.to_messages()
-        elim_response = await self.llm_main.ainvoke(elim_messages)
-        elim_output = getattr(elim_response, "content", elim_response)
-
-        # 3) learning_objective_finetuner (llm_main)
-        prompt_fine = await self.template_finetuner.aformat_prompt(
-            elimination_output=elim_output,
-            standardized_studytext=standardized_studytext
-        )
-        fine_messages = prompt_fine.to_messages()
-        fine_response = await self.llm_main.ainvoke(fine_messages)
-        fine_output = getattr(fine_response, "content", fine_response)
-
-        # 4) learning_objective_presenter (llm_main)
-        prompt_present = await self.template_presenter.aformat_prompt(
-            finetuned_output=fine_output,
-            standardized_studytext=standardized_studytext
-        )
-        present_messages = prompt_present.to_messages()
-        present_response = await self.llm_main.ainvoke(present_messages)
-        final_output = getattr(present_response, "content", present_response)
-
-        return final_output
-
     class Config:
         arbitrary_types_allowed = True

chains/learning_objectives_generator/runner.py

@@ -34,6 +34,8 @@ async def run_learning_objectives_generator(
     llm_a = llms.get(model_choice_1, config["default_llm_a"])
     llm_b = llms.get(model_choice_2, config["default_llm_b"])
 
+    llm_sanitize=llms.get(config["llm_sanitize"])
+
     # We will store the final sanitized results in an array of 4 strings
     # (2 prompts × 2 LLMs)
     partial_results = ["", "", "", ""]
@@ -50,7 +52,7 @@ async def run_learning_objectives_generator(
 
         # Step: sanitize
         sanitize_msg = await sanitize_prompt.aformat_prompt(raw_output=generation_output)
-        sanitize_resp = await llm_a.ainvoke(sanitize_msg.to_messages())  # or use a separate LLM for sanitization
+        sanitize_resp = await llm_sanitize.ainvoke(sanitize_msg.to_messages())  # or use a separate LLM for sanitization
         sanitized_output = getattr(sanitize_resp, "content", sanitize_resp)
 
         return (track_index, sanitized_output)

config/chain_configs.py

@@ -55,7 +55,11 @@ chain_configs = {
         "template_gen_prompt_a": template_gen_prompt_a,
         "template_gen_prompt_b": template_gen_prompt_b,
         "default_llm_a": llms["o1"],
-        "default_llm_b": llms["o3-mini (high reasoning_effort)"]
+        "default_llm_b": llms["o3-mini (high reasoning_effort)"],
+        "template_sanitize": template_sanitize_learning_objectives,
+        "llm_sanitize": llms["GPT-4o-mini (zero temp)"],
+
+
     },
 }
 

config/templates.py

@@ -368,3 +368,13 @@ The latter objective does not specify a single fact but combines two (can be pai
     ],
     input_variables=["standardized_text"]
 )
+
+template_sanitize_learning_objectives = ChatPromptTemplate(
+    messages=[
+        ("system", "You are given an output of a brainstorming session that lead to the generation of learning objectives. Your task is to "
+                   "turn distill this into a numbered list of just the learning objectives."),
+        ("human", "Here is the output:\n "
+                  "{raw_output}")
+    ],
+    input_variables=["raw_output"]
+)