Merge branch 'master' into harrison/use_output_parser

docs/examples/chains/llm_for_loop.ipynb

@@ -0,0 +1,290 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "4c475754",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.prompts import PromptTemplate\n",
+    "from langchain.prompts.base import BaseOutputParser\n",
+    "from langchain import OpenAI, LLMChain\n",
+    "from langchain.chains.llm_for_loop import LLMForLoopChain"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
+   "id": "efcdd239",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# First we make a chain that generates the list"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "id": "2b1884f5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing import Optional\n",
+    "import re\n",
+    "class ListOutputParser(BaseOutputParser):\n",
+    "    \n",
+    "    def __init__(self, regex: Optional[str] = None):\n",
+    "        self.regex=regex\n",
+    "    \n",
+    "    def parse(self, text: str) -> list:\n",
+    "        splits = [t for t in text.split(\"\\n\") if t]\n",
+    "        if self.regex is not None:\n",
+    "            splits = [re.match(self.regex, s).group(1) for s in splits]\n",
+    "        return splits"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 30,
+   "id": "b2b7f8fa",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "template = \"\"\"You are a list maker. Your job is make lists given a certain user input.\n",
+    "\n",
+    "The format of your lists should be:\n",
+    "\n",
+    "```\n",
+    "List:\n",
+    "- Item 1\n",
+    "- Item 2\n",
+    "...\n",
+    "```\n",
+    "\n",
+    "Begin!:\n",
+    "\n",
+    "User input: {input}\n",
+    "List:\"\"\"\n",
+    "output_parser = ListOutputParser(regex=\"- (.*)\")\n",
+    "prompt = PromptTemplate(template=template, input_variables=[\"input\"], output_parser=output_parser)\n",
+    "\n",
+    "chain = LLMChain(llm=OpenAI(temperature=0), prompt=prompt)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 44,
+   "id": "2f8ea6ba",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "['Tesla', 'Nissan', 'BMW', 'BYD', 'Volkswagen']"
+      ]
+     },
+     "execution_count": 44,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "chain.predict_and_parse(input=\"top 5 ev companies\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 32,
+   "id": "1fdfc7cb",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Next we generate the chain that we run over each item"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 38,
+   "id": "0b8f115a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "template = \"\"\"For the following company, explain the origin of their name:\n",
+    "\n",
+    "Company: {company}\n",
+    "Explanation of their name:\"\"\"\n",
+    "prompt = PromptTemplate(template=template, input_variables=[\"company\"])\n",
+    "\n",
+    "explanation_chain = LLMChain(llm=OpenAI(), prompt=prompt, verbose=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 39,
+   "id": "6d636881",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: Tesla\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "'\\n\\nTesla is a company that specializes in electric cars and renewable energy. The company is named after Nikola Tesla, a Serbian-American inventor and electrical engineer who was born in the 19th century.'"
+      ]
+     },
+     "execution_count": 39,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "explanation_chain.predict(company=\"Tesla\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 40,
+   "id": "3c236dd3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Now we combine them"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 41,
+   "id": "941b6389",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "for_loop_chain = LLMForLoopChain(llm_chain=chain, apply_chain=explanation_chain)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 45,
+   "id": "98c39dbc",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: Tesla\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n",
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: Nissan\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n",
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: BMW\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n",
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: BYD\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n",
+      "\n",
+      "\n",
+      "\u001b[1m> Entering new chain...\u001b[0m\n",
+      "Prompt after formatting:\n",
+      "\u001b[32;1m\u001b[1;3mFor the following company, explain the origin of their name:\n",
+      "\n",
+      "Company: Volkswagen\n",
+      "Explanation of their name:\u001b[0m\n",
+      "\n",
+      "\u001b[1m> Finished chain.\u001b[0m\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "['\\n\\nTesla was named after the Serbian-American inventor Nikola Tesla, who was known for his work in electricity and magnetism.',\n",
+       " '\\n\\nNissan is a Japanese company, and their name comes from the Japanese word for \"sun.\"',\n",
+       " \"\\n\\nThe company's name is an abbreviation for Bayerische Motoren Werke, which is German for Bavarian Motor Works.\",\n",
+       " '\\n\\nThe company\\'s name is derived from the Chinese characters \"Baiyu Dong\", which literally mean \"to catch the rain in the east\". The name is a reference to the company\\'s origins in the city of Shenzhen, in southeastern China.',\n",
+       " '\\n\\nVolkswagen is a German car company. The word \"Volkswagen\" means \"people\\'s car\" in German.']"
+      ]
+     },
+     "execution_count": 45,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "for_loop_chain.run_list(input=\"top 5 ev companies\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "5a2c1803",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

langchain/chains/llm.py

@@ -1,5 +1,5 @@
 """Chain that just formats a prompt and calls an LLM."""
-from typing import Any, Dict, List
+from typing import Any, Dict, List, Union
 
 from pydantic import BaseModel, Extra
 
@@ -78,3 +78,11 @@ class LLMChain(Chain, BaseModel):
                 completion = llm.predict(adjective="funny")
         """
         return self(kwargs)[self.output_key]
+
+    def predict_and_parse(self, **kwargs: Any) -> Union[str, List[str], Dict[str, str]]:
+        """Call predict and then parse the results."""
+        result = self.predict(**kwargs)
+        if self.prompt.output_parser is not None:
+            return self.prompt.output_parser.parse(result)
+        else:
+            return result

langchain/chains/llm_for_loop.py

@@ -0,0 +1,63 @@
+"""Chain that first uses an LLM to generate multiple items then loops over them."""
+from typing import Any, Dict, List
+
+from pydantic import BaseModel, Extra, root_validator
+
+from langchain.chains.base import Chain
+from langchain.chains.llm import LLMChain
+from langchain.prompts.base import ListOutputParser
+
+
+class LLMForLoopChain(Chain, BaseModel):
+    """Chain that first uses an LLM to generate multiple items then loops over them."""
+
+    llm_chain: LLMChain
+    """LLM chain to use to generate multiple items."""
+    apply_chain: Chain
+    """Chain to apply to each item that is generated."""
+    output_key: str = "text"  #: :meta private:
+
+    class Config:
+        """Configuration for this pydantic object."""
+
+        extra = Extra.forbid
+        arbitrary_types_allowed = True
+
+    @property
+    def input_keys(self) -> List[str]:
+        """Will be whatever keys the prompt expects.
+
+        :meta private:
+        """
+        return self.llm_chain.prompt.input_variables
+
+    @property
+    def output_keys(self) -> List[str]:
+        """Will always return text key.
+
+        :meta private:
+        """
+        return [self.output_key]
+
+    @root_validator()
+    def validate_output_parser(cls, values: Dict) -> Dict:
+        """Validate that the correct inputs exist for all chains."""
+        chain = values["llm_chain"]
+        if not isinstance(chain.prompt.output_parser, ListOutputParser):
+            raise ValueError(
+                f"The OutputParser on the base prompt should be of type "
+                f"ListOutputParser, got {type(chain.prompt.output_parser)}"
+            )
+        return values
+
+    def run_list(self, **kwargs: Any) -> List[str]:
+        """Get list from LLM chain and then run chain on each item."""
+        output_items = self.llm_chain.predict_and_parse(**kwargs)
+        res = []
+        for item in output_items:
+            res.append(self.apply_chain.run(item))
+        return res
+
+    def _call(self, inputs: Dict[str, str]) -> Dict[str, str]:
+        res = self.run_list(**inputs)
+        return {self.output_key: "\n\n".join(res)}

langchain/prompts/base.py

@@ -2,10 +2,10 @@
 import json
 from abc import ABC, abstractmethod
 from pathlib import Path
-from typing import Any, Dict, List, Union
+from typing import Any, Dict, List, Optional, Union
 
 import yaml
-from pydantic import BaseModel, root_validator
+from pydantic import BaseModel, Extra, root_validator
 
 from langchain.formatting import formatter
 
@@ -32,11 +32,35 @@ def check_valid_template(
         raise ValueError("Invalid prompt schema.")
 
 
+class BaseOutputParser(ABC):
+    """Class to parse the output of an LLM call."""
+
+    @abstractmethod
+    def parse(self, text: str) -> Union[str, List[str], Dict[str, str]]:
+        """Parse the output of an LLM call."""
+
+
+class ListOutputParser(ABC):
+    """Class to parse the output of an LLM call to a list."""
+
+    @abstractmethod
+    def parse(self, text: str) -> List[str]:
+        """Parse the output of an LLM call."""
+
+
 class BasePromptTemplate(BaseModel, ABC):
     """Base prompt should expose the format method, returning a prompt."""
 
     input_variables: List[str]
     """A list of the names of the variables the prompt template expects."""
+    output_parser: Optional[BaseOutputParser] = None
+    """How to parse the output of calling an LLM on this formatted prompt."""
+
+    class Config:
+        """Configuration for this pydantic object."""
+
+        extra = Extra.forbid
+        arbitrary_types_allowed = True
 
     @root_validator()
     def validate_variable_names(cls, values: Dict) -> Dict: