A simple Python implementation of the ReAct pattern for LLMs

Simon Willison published a from-scratch Python implementation of the ReAct (Reason+Act) pattern on top of his tiny ChatGPT API wrapper, giving the LLM three callable actions: `wikipedia: <search term>`, `simon_blog_search: <search term>`, and `calculate: <expression>`, the last evaluated with Python's `eval()` — which Willison flagged as "so dangerous!" and said should use a safer mechanism. In worked examples, the model answered "England shares borders with Wales and Scotland" via a Wikipedia lookup and resolved "Fifteen * twenty five" by requesting the calculate action. Willison built the implementation himself rather than using langchain, which Matt Webb had used for his version of the loop.

A popular nightmare scenario for AI is giving it access to tools, so it can make API calls and execute its own code and generally break free of the constraints of its initial environment. Let's do that now The ReAct pattern for Reason+Act is described in this paper https://react-lm.github.io/ . It's a pattern where you implement additional actions that an LLM can take - searching Wikipedia or running calculations for example - and then teach it how to request that those actions are run, then feed their results back into the LLM. See also my post Could you train a ChatGPT-beating model for $85,000 and run it in a browser? https://simonwillison.net/2023/Mar/17/beat-chatgpt-in-a-browser/ react-pattern . Matt Webb wrote a great piece about that here: The surprising ease and effectiveness of AI in a loop https://interconnected.org/home/2023/03/16/singularity . In it Matt points out that the 1/10 price reduction on ChatGPT compared to GPT-3 makes it an ideal candidate for cheaply implementing this pattern. Matt used langchain https://github.com/hwchase17/langchain for his implementation. I decided to try building it from scratch on top of my tiny Python wrapper https://til.simonwillison.net/gpt3/chatgpt-api for the ChatGPT API. Here's my first attempt at a very rough implementation. It gains access to three new actions: wikipedia: <search term - search Wikipedia and returns the snippet of the first result simon blog search: <search term - search my blog and return the truncated text of the first result calculate: <expression - evaluate an expression using Python's eval function so dangerous This should absolutely use something like Here are some very cherry-picked examples of my query function running, showing the underlying deliberations of the LLM: query "What does England share borders with?" Thought: I should list down the neighboring countries of England Action: wikipedia: England PAUSE -- running wikipedia England Observation: <span class="searchmatch" England</span is a country that is part of the United Kingdom. It shares land borders with Wales to its west and Scotland to its north. The Irish Sea lies northwest Answer: England shares borders with Wales and Scotland. query "Has Simon been to Madagascar?" Thought: I can try and search on Simon's blog or look up Madagascar on Wikipedia to see if Simon has mentioned it in any of his writings. Action: simon blog search: Madagascar PAUSE -- running simon blog search Madagascar Observation: Weeknotes: More releases, more museums: Lots of small releases this week. Datasette I released two bug fix releases for Datasette - 0.30.1 and 0.30.2. Changelog here. My Dogsheep personal analytics project means I& 39;m using Datasette for my own data analysis every day, which inspires me to fix small but annoying bugs much more aggressively. I& 39;ve also set myself a Streak goal to land a commit to Datasette every day. I landed a tiny new feature to master yesterday: a ?column notin=x,y,z filter, working as an inverse of the existing ?column in=x,y,z filter. See issue 614 for details. More Niche Museums I& 39;ve been keeping up my streak of adding at least one new museum to www.niche-museums.com every day. This week I added the Pirates Museum in Antananarivo, Madagascar, the David Rumsey Map Center at Stanford, Galerie de Paléontologie et d’Anatomie comparée in Paris, DEVIL-ish Little Things in Vancouver, Washington, Mardi Gras World in New Orleans, Environmental Volunteers EcoCenter in Palo Alto, the Evergreen A Observation: According to the blog post searched on Simon's blog, he has visited the Pirates Museum in Antananarivo, Madagascar which is listed as a niche museum on www.niche-museums.com. Answer: Yes, Simon has been to Madagascar and visited the Pirates Museum located in Antananarivo. query "Fifteen twenty five" Thought: The action required is a calculation Action: calculate: 15 25 PAUSE -- running calculate 15 25 Observation: 375 Answer: Fifteen times twenty five equals 375. This code is Apache 2 licensed: https://www.apache.org/licenses/LICENSE-2.0 import openai import re import httpx openai.api key = "sk-..." class ChatBot: def init self, system="" : self.system = system self.messages = if self.system: self.messages.append {"role": "system", "content": system} def call self, message : self.messages.append {"role": "user", "content": message} result = self.execute self.messages.append {"role": "assistant", "content": result} return result def execute self : completion = openai.ChatCompletion.create model="gpt-3.5-turbo", messages=self.messages Uncomment this to print out token usage each time, e.g. {"completion tokens": 86, "prompt tokens": 26, "total tokens": 112} print completion.usage return completion.choices 0 .message.content prompt = """ You run in a loop of Thought, Action, PAUSE, Observation. At the end of the loop you output an Answer Use Thought to describe your thoughts about the question you have been asked. Use Action to run one of the actions available to you - then return PAUSE. Observation will be the result of running those actions. Your available actions are: calculate: e.g. calculate: 4 7 / 3 Runs a calculation and returns the number - uses Python so be sure to use floating point syntax if necessary wikipedia: e.g. wikipedia: Django Returns a summary from searching Wikipedia simon blog search: e.g. simon blog search: Django Search Simon's blog for that term Always look things up on Wikipedia if you have the opportunity to do so. Example session: Question: What is the capital of France? Thought: I should look up France on Wikipedia Action: wikipedia: France PAUSE You will be called again with this: Observation: France is a country. The capital is Paris. You then output: Answer: The capital of France is Paris """.strip action re = re.compile '^Action: \w+ : . $' def query question, max turns=5 : i = 0 bot = ChatBot prompt next prompt = question while i < max turns: i += 1 result = bot next prompt print result actions = action re.match a for a in result.split '\n' if action re.match a if actions: There is an action to run action, action input = actions 0 .groups if action not in known actions: raise Exception "Unknown action: {}: {}".format action, action input print " -- running {} {}".format action, action input observation = known actions action action input print "Observation:", observation next prompt = "Observation: {}".format observation else: return def wikipedia q : return httpx.get "https://en.wikipedia.org/w/api.php", params={ "action": "query", "list": "search", "srsearch": q, "format": "json" } .json "query" "search" 0 "snippet" def simon blog search q : results = httpx.get "https://datasette.simonwillison.net/simonwillisonblog.json", params={ "sql": """ select blog entry.title || ': ' || substr html strip tags blog entry.body , 0, 1000 as text, blog entry.created from blog entry join blog entry fts on blog entry.rowid = blog entry fts.rowid where blog entry fts match escape fts :q order by blog entry fts.rank limit 1""".strip , " shape": "array", "q": q, } .json return results 0 "text" def calculate what : return eval what known actions = { "wikipedia": wikipedia, "calculate": calculate, "simon blog search": simon blog search } This is not a very robust implementation at all - there's a ton of room for improvement. But I love how simple it is - it really does just take a few dozen lines of Python to make these extra capabilities available to the LLM and have it start to use them. Created 2023-03-17T07:52:22-07:00, updated 2023-03-20T15:18:43-07:00 · History https://github.com/simonw/til/commits/main/llms/python-react-pattern.md · Edit https://github.com/simonw/til/blob/main/llms/python-react-pattern.md