Heya 👋
We're back with Quine's #TrendingTuesday. This week we created a list for the fastest-growing repos in Python!
It's really incredible to see that nine out of ten repos have to do with LLMs and Machine-Learning 🤯
🔟 In place number 10 we have ComfyUI, a user interface that allows users to create and execute complex Stable Diffusion pipelines without needing to write code!
It has a nodes/graph/flowchart interface that enables users to experiment with different workflows. It also works fully offline and starts up quickly!
9️⃣ In number nine, we have a repo by Microsoft.
LMOps is a dissemination repo for a new initiative that focuses on fundamental research and technology for building products that use LLMs and Generative AI models.
The initiative has a focus on developing tools for generating better prompts, developing methods for longer input contexts, and on theoretical foundations of LLMs.
8️⃣ Next is TensorLayerX which is an AI framework that allows users to run their AI code on different types of hardware such as Nvidia-GPU, Huawei-Ascend, and Cambricon. It also supports multiple backends including TensorFlow, PyTorch, MindSpore, PaddlePaddle, OneFlow, and Jittor.
The project is maintained by peeps from Peking University, Imperial College London, Princeton, Stanford, and a few other institutions!
7️⃣ text-generation-webui allows you to summon the power of the main LLMs from the comfort of your browser!
The project aims to be the leading UI to interact with LLMs via a web browser. Its interface was built with Gradio and supports GPT-J 6B, OPT, GALACTICA, GPT-Neo, and Pygmalion.
Users can switch between different models using a dropdown menu and generate text by entering prompts or questions 🙋
6️⃣ In place number six is the only non-AI repo of the list :)
NimPlant is an offensive security tool that lets "hackers" perform operations on a target machine. It encrypts and compresses all traffic to and from the target machine and supports a wide range of command-line like commands that can be used to perform early-stage operations like gathering information about the machine or interacting with files and the registry.
It's primarily written in the Nim programming language (22.8%), but it is also 20% Python.
5️⃣ In number five, we have a very fun application of LLMs and GPT-like technologies...
shyamsn97/mario-gpt is the official repo for MarioGPT - a GPT2-based model trained to generate game levels for Super Mario Bros! 🍄👸
If you're into LLMs, we highly recommend you look into the associated arxiv paper where the authors go into the details about Procedural Content Generation and novelty search.
4️⃣ Another GPT-related repo ❤️
picoGPT is a really small implementation of GPT-2, made entirely in NumPy.
It's so small that the forward pass code is only 40 lines long! While it's not very fast and doesn't have any training code or support for batch inference, it's incredibly easy to read. Check it out if you're interested in learning how GPT-2 works ✨
3️⃣ Mikubill/sd-webui-controlnet is a work-in-progress extension for AUTOMATIC1111's Stable Diffusion web UI.
It enables the addition of ControlNet, a neural network that allows to generate more customised outputs (see next repo for more details 😉). The project is not yet fully functional, but is quickly gaining support from the community!
2️⃣ In second place we have lllyasviel/ControlNet which is the official implementation of ControlNet, a neural network structure that allows you to use pre-trained large diffusion models with additional input conditions, such as edge and segmentation maps.
ControlNet learns the specific conditions of a task in an end-to-end way, and can do so even if the training dataset is small (< 50k). Check out its arxiv paper for more details.
1️⃣ The fastest growing Python repo of the week is mm-cot 🏆
This repo contains the official implementation of the Multimodal-CoT model which combines both text and image data to generate intermediate reasoning chains to better infer answers. This model has less than 1 billion parameters and shows better performance than GPT-3.5 on some tasks!
Just like your books, this model is brought to you by Amazon
We'll tweet about trending GitHub repos every week!
Did you like this list? Give us a follow here and on Twitter 🐦
Top comments (3)
Very informative. Definitive appropriate here!
Some comments may only be visible to logged-in visitors. Sign in to view all comments.