track intermediate steps

docs/examples/agents/self_ask_with_search.ipynb

@@ -13,6 +13,26 @@
   {
    "cell_type": "code",
    "execution_count": 1,
+   "id": "51cbc105",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain import OpenAI, SerpAPIWrapper\n",
+    "from langchain.agents import initialize_agent, Tool\n",
+    "\n",
+    "llm = OpenAI(temperature=0)\n",
+    "search = SerpAPIWrapper()\n",
+    "tools = [\n",
+    "    Tool(\n",
+    "        name=\"Intermediate Answer\",\n",
+    "        func=search.run\n",
+    "    )\n",
+    "]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
    "id": "7e3b513e",
    "metadata": {},
    "outputs": [
@@ -39,24 +59,12 @@
        "'El Palmar, Spain'"
       ]
      },
-     "execution_count": 1,
+     "execution_count": 2,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "from langchain import OpenAI, SerpAPIWrapper\n",
-    "from langchain.agents import initialize_agent, Tool\n",
-    "\n",
-    "llm = OpenAI(temperature=0)\n",
-    "search = SerpAPIWrapper()\n",
-    "tools = [\n",
-    "    Tool(\n",
-    "        name=\"Intermediate Answer\",\n",
-    "        func=search.run\n",
-    "    )\n",
-    "]\n",
-    "\n",
     "self_ask_with_search = initialize_agent(tools, llm, agent=\"self-ask-with-search\", verbose=True)\n",
     "\n",
     "self_ask_with_search.run(\"What is the hometown of the reigning men's U.S. Open champion?\")"
@@ -64,9 +72,51 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 4,
    "id": "683d69e7",
    "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new SelfAskWithSearchAgent chain...\u001b[0m\n",
+      "What is the hometown of the reigning men's U.S. Open champion?\n",
+      "Are follow up questions needed here:\u001b[32;1m\u001b[1;3m Yes.\n",
+      "Follow up: Who is the reigning men's U.S. Open champion?\u001b[0m\n",
+      "Intermediate answer: \u001b[36;1m\u001b[1;3mCarlos Alcaraz\u001b[0m\n",
+      "\u001b[32;1m\u001b[1;3mFollow up: Where is Carlos Alcaraz from?\u001b[0m\n",
+      "Intermediate answer: \u001b[36;1m\u001b[1;3mEl Palmar, Spain\u001b[0m\n",
+      "\u001b[32;1m\u001b[1;3mSo the final answer is: El Palmar, Spain\u001b[0m\n",
+      "\u001b[1m> Finished SelfAskWithSearchAgent chain.\u001b[0m\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "{'input': \"What is the hometown of the reigning men's U.S. Open champion?\",\n",
+       " 'output': 'El Palmar, Spain',\n",
+       " 'intermediate_steps': \"What is the hometown of the reigning men's U.S. Open champion?\\nAre follow up questions needed here: Yes.\\nFollow up: Who is the reigning men's U.S. Open champion?\\nIntermediate answer: Carlos Alcaraz\\nFollow up: Where is Carlos Alcaraz from?\\nIntermediate answer: El Palmar, Spain\\nSo the final answer is: El Palmar, Spain\"}"
+      ]
+     },
+     "execution_count": 4,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "self_ask_with_search = initialize_agent(tools, llm, agent=\"self-ask-with-search\", verbose=True, return_intermediate_steps=True)\n",
+    "\n",
+    "self_ask_with_search({\"input\": \"What is the hometown of the reigning men's U.S. Open champion?\"})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b30a1b9a",
+   "metadata": {},
    "outputs": [],
    "source": []
   }

langchain/agents/agent.py

@@ -26,6 +26,8 @@ class Agent(Chain, BaseModel, ABC):
     prompt: ClassVar[BasePromptTemplate]
     llm_chain: LLMChain
     tools: List[Tool]
+    return_intermediate_steps: bool = False
+    intermediate_steps_key: str = "intermediate_steps"  #: :meta private:
     input_key: str = "input"  #: :meta private:
     output_key: str = "output"  #: :meta private:
 
@@ -43,7 +45,10 @@ class Agent(Chain, BaseModel, ABC):
 
         :meta private:
         """
-        return [self.output_key]
+        if self.return_intermediate_steps:
+            return [self.output_key, self.intermediate_steps_key]
+        else:
+            return [self.output_key]
 
     @property
     @abstractmethod
@@ -146,7 +151,10 @@ class Agent(Chain, BaseModel, ABC):
             chained_input.add(output.log, color="green")
             # If the tool chosen is the finishing tool, then we end and return.
             if output.tool == self.finish_tool_name:
-                return {self.output_key: output.tool_input}
+                final_output = {self.output_key: output.tool_input}
+                if self.return_intermediate_steps:
+                    final_output[self.intermediate_steps_key] = chained_input.input
+                return final_output
             # Otherwise we lookup the tool
             chain = name_to_tool_map[output.tool]
             # We then call the tool on the tool input to get an observation