Add documentation for custom prompts for Agents (#631) (#640)

- Added a comment interpreting regex for `ZeroShotAgent`
- Added a note to the `Custom Agent` notebook

Co-authored-by: Sam Ching <samuel@duolingo.com>

docs/modules/agents/examples/custom_agent.ipynb

@@ -133,6 +133,17 @@
     "print(prompt.template)"
    ]
   },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "id": "5e028e6d",
+   "metadata": {},
+   "source": [
+    "Note that we are able to feed agents a self-defined prompt template, i.e. not restricted to the prompt generated by the `create_prompt` function, assuming it meets the agent's requirements. \n",
+    "\n",
+    "For example, for `ZeroShotAgent`, we will need to ensure that it meets the following requirements. There should a string starting with \"Action:\" and a following string starting with \"Action Input:\", and both should be separated by a newline.\n"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 16,
@@ -350,7 +361,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3.9.0 64-bit ('llm-env')",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -364,11 +375,11 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.0"
+   "version": "3.8.12 (default, Feb 15 2022, 17:41:09) \n[Clang 12.0.5 (clang-1205.0.22.11)]"
   },
   "vscode": {
    "interpreter": {
-    "hash": "b1677b440931f40d89ef8be7bf03acb108ce003de0ac9b18e8d43753ea2e7103"
+    "hash": "18784188d7ecd866c0586ac068b02361a6896dc3a29b64f5cc957f09c590acef"
    }
   }
  },

langchain/agents/mrkl/base.py

@@ -28,7 +28,13 @@ class ChainConfig(NamedTuple):
 
 
 def get_action_and_input(llm_output: str) -> Tuple[str, str]:
-    """Parse out the action and input from the LLM output."""
+    """Parse out the action and input from the LLM output.
+
+    Note: if you're specifying a custom prompt for the ZeroShotAgent,
+    you will need to ensure that it meets the following Regex requirements.
+    The string starting with "Action:" and the following string starting
+    with "Action Input:" should be separated by a newline.
+    """
     if FINAL_ANSWER_ACTION in llm_output:
         return "Final Answer", llm_output.split(FINAL_ANSWER_ACTION)[-1].strip()
     regex = r"Action: (.*?)\nAction Input: (.*)"