tree-of-thought-llm
pythagora
tree-of-thought-llm | pythagora | |
---|---|---|
41 | 37 | |
4,228 | 1,520 | |
4.3% | 2.8% | |
7.2 | 7.8 | |
3 months ago | 5 days ago | |
Python | JavaScript | |
MIT License | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
tree-of-thought-llm
-
AI Chat Applications with the Metacognition Approach: Tree of Thoughts (ToT)
[2305.10601] Tree of Thoughts: Deliberate Problem Solving with Large Language Models (arxiv.org)
-
Last night /u/ alesneolith posted a very serious writeup claiming to have worked in one of the projects. The writeup is more elaborate than expected and got surprisingly little attention. His account has been since deleted.
Language models are increasingly being deployed for general problem solving across a wide range of tasks, but are still confined to token-level, left-to-right decision-making processes during inference. This means they can fall short in tasks that require exploration, strategic lookahead, or where initial decisions play a pivotal role. To surmount these challenges, we introduce a new framework for language model inference, “Tree of Thoughts” (ToT), which generalizes over the popular “Chain of Thought” approach to prompting language models, and enables exploration over coherent units of text (“thoughts”) that serve as intermediate steps toward problem solving. ToT allows LMs to perform deliberate decision making by considering multiple different reasoning paths and self-evaluating choices to decide the next course of action, as well as looking ahead or backtracking when necessary to make global choices. Our experiments show that ToT significantly enhances language models’ problem-solving abilities on three novel tasks requiring non-trivial planning or search: Game of 24, Creative Writing, and Mini Crosswords. For instance, in Game of 24, while GPT-4 with chain-of-thought prompting only solved 4% of tasks, our method achieved a success rate of 74%. Code repo with all prompts: https://github.com/princeton-nlp/tree-of-thought-llm.
-
Ultra Fast Bert
GPU utilization should be down when using this technique. I’m hoping this could allow for more efficient batch inference on GPUs. If you can predict 10 tokens for the price of 1 it should allow you to do tree of thought much more efficiently.
https://github.com/princeton-nlp/tree-of-thought-llm
-
Is it best to not pay attention to AI news and/or find ways to delude ourselves into believing better outcomes?
For those familiar with Daniel Kahneman's Thinking Fast and Slow, the current LLMs (such as GPT-4 via ChatGPT) seem to resemble System 1 thinking (near-instantaneous, automatic, intuitive processes like next-word prediction). However, they lack System 2 thinking (slow, effortful, logical, planning, reasoning). What I learned today is that Google's Gemini (an LLM in training now) not only has more modalities (I think all Youtube Video and audio??), more compute, and almost twice the training data, but they're building in AlphaGo-type learning, which resembles tree of thoughts and looks a LOT like the missing puzzle piece of System 2 thinking. Will it be AGI? Maybe, and it's coming this winter.
-
Langchain Is Pointless
Tree of thoughts: https://arxiv.org/abs/2305.10601
Good video on "Tree of thoughts" which also reviews / puts it in the context of other methods: https://www.youtube.com/watch?v=ut5kp56wW_4
Completion vs conversational interface is something you can read about in the OpenAI API documentation.
For the remaining things I don't have single specific pointer at hand.
-
To all skeptics with a background in AI/CS : what is your realistic timeline for AGI/ASI ?
What do you think about the combination of Tree of Thoughts: Deliberate Problem Solving with Large Language Models LongNet: Scaling Transformers to 1,000,000,000 Tokens Textbooks Are All You Need Attention Is All You Need Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
-
Why do language models appear to work left-to-right?
You are right. Tree of Thoughts: Deliberate Problem Solving with Large Language Models proposes to solve this via MCTS-style generation (similar to how AlphaGo worked, and a lot of planning & control problems are executed).
-
Munk Debate on Artificial Intelligence
The transformer was developed in 2017 and it powers all modern LLMs. If you're familiar with Daniel Kahneman's work from Thinking Fast and Slow, you could easily summarize LLMs as excellent System 1 thinking: our fast, automatic, unconscious responses (e.g. autocomplete). I'd argue that we're one development (similar to the transformer) away from creating System 2 thinking: deliberate and strategic thinking. In fact, with merely GPT-4 and some clever architectures, researchers have developed chain-of-thought prompting and, more recently, tree-of-thoughts reasoning. While external to the LLM architecture, embedding these concepts into a LLM could very likely solve the creation of System 2 thinking and produce the first real AGI. Adding more modalities (e.g. audio, images, video, topography, etc.) will simply add more nuance in the weights and biases of a complete system.
-
Question regarding model compatibility for Alpaca Turbo
There are a bunch of other methods to improve quality and performance like tree-of-thought-llm, connecting a LLM to a database or have it review its own output.
- Tree of thoughts build in open-source model
pythagora
-
AI Chat Applications with the Metacognition Approach: Tree of Thoughts (ToT)
Product which you can try - https://github.com/Pythagora-io/pythagora, check also video - Open-Source AI Agent Can Build FULL STACK Apps (FREE “Devin” Alternative) (youtube.com)
-
How to kickstart automated test suite when there are 0 tests written and the codebase is already huge
P.S. If you found this post helpful, it would mean a lot to me if you starred the Pythagora Github repo and if you try Pythagora out, please let us know how it went on [email protected].
- Show HN: CLI tool that writes unit tests for Node.js apps with GPT-4
-
I created a CLI tool that writes unit tests with GPT-4 (with one command, I created tests for Lodash repo with 90% code coverage and found 13 bugs)
Thanks, yes, it can, we actually started off with integration tests. Take a look at the integration tests README. They work by recording server activity (db queries, 3rd party API requests, etc.) during the processing of an API request.
- Pythagora creates automated tests for you by analysing server activity
-
I created a dev tool that uses GPT-4 to generate integration tests in Jest by tracking server activity
Recently, I open sourced Pythagora - a dev tool that tracks the server activity and creates integration tests from it. However, many people said that they wanted to analyze better what is inside the tests and to use tests as code documentation. So, I used GPT-4 to export Pythagora tests (which are basically JSON files that contain all captured data) to Jest code which can be reviewed and used for documentation.
-
45 ways to break an API server (negative tests with examples)
I'm working on Pythagora, an open source tool that writes automated integration tests by itself (well, with a bit of help from GPT-4) without you, the dev, having to write a single line of code. Basically, you can get from 0 to 80% code code coverage within 30 minutes (video).
-
What project are you currently working on?
Hey! I'm working on Pythagora (https://github.com/Pythagora-io/pythagora) - it's an open source tool that creates automated integration tests by analyzing server activity without you having to write a single line of code.
-
How Pythagora Reduces Debugging Time and Supercharges Your Development Workflow
Pythagora is an NPM package designed to create automated tests for your Node.js applications by analyzing server activity. It records all requests to your app's endpoints, along with responses and server actions, such as Mongo and Redis queries. Pythagora during testing simulates the server conditions from the time when the request was captured, allowing for consistent and accurate testing across different environments.
-
Creating integration tests for a backend legacy codebase
Finally, if you read this far and would like to support us, please consider starring the Pythagora Github repository here – it would mean the world to us.
What are some alternatives?
Voyager - An Open-Ended Embodied Agent with Large Language Models
mern-ecommerce - :balloon: Fullstack MERN Ecommerce Application
guidance - A guidance language for controlling large language models. [Moved to: https://github.com/guidance-ai/guidance]
fastkafka - FastKafka is a powerful and easy-to-use Python library for building asynchronous web services that interact with Kafka topics. Built on top of Pydantic, AIOKafka and AsyncAPI, FastKafka simplifies the process of writing producers and consumers for Kafka topics.
tree-of-thoughts - Plug in and Play Implementation of Tree of Thoughts: Deliberate Problem Solving with Large Language Models that Elevates Model Reasoning by atleast 70%
change-case - Convert strings between camelCase, PascalCase, Capital Case, snake_case and more
Neurite - Fractal Graph Desktop for Ai-Agents, Web-Browsing, Note-Taking, and Code.
js-proper-url-join - Like path.join but for a URL
hamilton - Hamilton helps data scientists and engineers define testable, modular, self-documenting dataflows, that encode lineage and metadata. Runs and scales everywhere python does.
api
Mr.-Ranedeer-AI-Tutor - A GPT-4 AI Tutor Prompt for customizable personalized learning experiences.
pythagora-demo-lodash - A modern JavaScript utility library delivering modularity, performance, & extras.