Refact is an AI coding assistant that combines code autocompletion, refactoring, and chat inside your favorite IDE. It offers a wholesome developer experience by making multiple functions available in one IDE, including autocomplete, code improvement, bug fixing, code explanation, and integrated AI chat. Refact supports multiple programming languages and provides models like GPT-4, Code LLama, StarCoder, and WizardCoder. It prioritizes privacy and offers self-hosted versions for those who prefer not to share their code.

Refact is an open-source AI code assistant with fine-tuning.

Refact.ai

Recently, we published a massive update of the Refact AI VS Code plugin.
Our toolbox has undergone a complete makeover! 

Check out the video to get a sneak peek of the new plugin!
https://lnkd.in/dnK_f3h6

📣Join our first livestream tomorrow on Discord at 5pm CET/ 11am ET with our Developer Advocate Vadim Smirnov.

 ⭐️⭐We will celebrate our recent milestone of 1k stars on GitHub and give a sneak peek of things we’re working on! 

Check the link to register in the comments.

🥳Today we're excited to share the Enterprise Refact edition that we've been working on tirelessly for the past couple of months.

Enterprise Refact will be perfect for teams who want to get local code completions tailored to their codebase and get the fastest inference possible. We have support for: 
- fine-tuning with Refact 1.6B model 
- vLLM for faster inference 
- load balancing between GPUs

With its easy set-up and intuitive interface, you can set up a cadence for personalized fine-tunes to adapt to your coding styles and get tailored recommendations. No special knowledge of LLMs is needed!

I am super excited to publish a video about fine-tuning with Refact AI!

It was so awesome to see that the accuracy of suggestions was significantly improved for Prismic and #nextjs!

https://lnkd.in/dWnYTYVM

🚀Today we're launching Refact code LLM! 

⭐️ It's 1.6B code model that reaches 32% HumanEval. In addition to blazing-fast fill-in-the-middle code completion, it can chat about code.

⭐️ The model was pre-trained on permissive data in 20+ programming languages using 1.2T tokens. It was then fine-tuned with instruction-following data using 40B tokens to improve performance.

⭐️ In training the model, we applied LLama, Early Dropout, Multi-Query Attention, ALiBi Position Encoding, and LiON innovations.

⭐️ We release Refact LLM under an OpenRAIL license agreement. It’s available for commercial use.

⭐️ It's available in the self-hosted version and it we'll soon be available in the cloud, too. Head to the links in the comments!

Huge announcement today:  Refact AI is now open-source and available to everyone🎉 

Here’re some of the new things that you can try …

🎯Local Fine-tuning 
You can now fine-tune a code LLM to your codebase locally.. Read more about our fine-tuning and how it can generate better suggestions than a base code model. 

🌟More powerful autocompletion, chat & support for new LLMs: 
We've added new LLMs in the self-hosted version to make autocompletion even more powerful. In addition to Refact and StarCoder models, we now have WizardCoder for code completion, and StarChat, WizardLM, Llama2 for a local chat.

💡If you have a feature idea or need support for more plugins, we're all ears! Share your thoughts with us on GitHub and join our Discord community.
We’re working on a bunch of new features, including context-aware chat on the entire codebase with vecDB, enterprise and team plans, and our next-gen code model.

Are you interested in training code models? 🤖

We just published a blog post sharing the principles and methods we use to train our latest next-gen model, check it out. 

https://lnkd.in/eAZcGMbs #code #llm

Refact.ai has joined the NVIDIA Inception program for startups 🎉. 

We’re excited to work together with the NVIDIA AI team to accelerate our mission of training next-gen generative AI models available for everyone. 

Stay tuned! 

#NVIDIAInception #AI #llm

💥Announcing Refact's new Self-Hosted version powered by the StarCoder open-source 15b code model. 

Refact users can now use code completion, code transformation, and chat functions powered by the high-performance StarCoder 15b model on their own GPU. 

With over 8,000 tokens input size in StarCoder it can use more context than any open LLM. This allows for a bunch of use cases from autocompletion, code refactoring, fixing bugs, to annotating and writing code from an instruction.

We're committed to continuing to support BigCode's mission of developing AI code models in an open and responsible way and empowering every developer to become more productive with AI.

Perhaps, the easiest way to make your code shorter: press F1 in Refact and relax while the magic is happening 🪄

AI:REFACT

Refact

Analysis Not Available