Astra Clelia Bertelli PRO
AI & ML interests
Articles
Organizations
as-cle-bert's activity
In the past days, OpenAI announced their search engine, SearchGPT: today, I'm glad to introduce you SearchPhi, an AI-powered and open-source web search tool that aims to reproduce similar features to SearchGPT, built upon microsoft/Phi-3-mini-4k-instruct, llama.cpp๐ฆ and Streamlit.
Although not as capable as SearchGPT, SearchPhi v0.0-beta.0 is a first step toward a fully functional and multimodal search engine :)
If you want to know more, head over to the GitHub repository (https://github.com/AstraBert/SearchPhi) and, to test it out, use this HF space: as-cle-bert/SearchPhi
Have fun!๐ฑ
Hope y'all are as excited as me for the release of Llama 3.1! ๐ฆ
Following the release, I built a space exploiting HF Inference API, thanks to a recipe you can find in this awesome GitHub repo (https://github.com/huggingface/huggingface-llama-recipes/): you can now run Llama-3.1-405B customizing its system instructions and other parameters, for free! ๐
Follow this link: as-cle-bert/Llama-3.1-405B-FP8 and let the fun begin!๐
Good news concerning as-cle-bert/smolLM-arena, the chat arena where you can compare some of the Small Language Models (<1.7B) on the Hub and cast your vote to choose the best!๐ฑ
The space now has a new interface with chatbots instead of textboxs, it runs faster and it also comes with usage instructions :)
Have fun!๐
The SmolLM series is specifically designed to run on devices like smartphones, yes :) And, concerning the arena for models 7 to 20B, I didn't want to spoiler it, but It's coming soon! ;)
Thanks for pointing that out, now it should work!
As you may know, small language models like the SmolLM series have been on the rise recently: although it may not be completely fair to compare them with larger models, my thought was that we could build a space where these SLMs could compete against each other in a chat arena, and here is what came out: as-cle-bert/smolLM-arena ๐
Even though there might be some little issues and hiccups due to GPU resources allocation, this space offers you the possibility to compare and play around with several Small Language Models, coming also with a leaderboard page (make sure to refresh it for the latest updates!)๐
Have fun!๐
If you're interested in the subject, you can check out my latest community article: https://huggingface.co/blog/as-cle-bert/is-ai-carbon-footprint-worrisome
Where I try to unravel AI's carbon footprint and potential solutions to reduce it๐ป
Enjoy!๐ค
There's a new space out in the wild! as-cle-bert/self-reviewing-coding-assistant ๐ฆ
It's a self-correcting and self-reviewing python coding assistant based on GPT4-o and LangChain, inspired by Codium-AI's AlphaCodium ๐พ
Have fun! ๐
๐ค A vital question that every developer may have asked themselves in the last three years is: how can we improve AI code generation?
๐ In my last Community blog post, I talk about Codium AI's AlphaCodium and how they tried to enhance LLMs coding skills with flow engineering: https://huggingface.co/blog/as-cle-bert/repetita-iuvant-how-to-improve-ai-code-generation
๐ Enjoy!
๐งฌ As you may now, Evolutionary Scale recently released EvolutionaryScale/esm3-sm-open-v1 model here on the Hub, "a frontier generative model for biology, able to jointly reason across three fundamental biological properties of proteins: sequence, structure, and function" - as it is described on the dedicated GitHub page.
โก If you are curious about it and you want to try it out, you can do it with a space I built, as-cle-bert/proteins-with-esm
Hope this helps with your research!๐
๐ฅ If you are Bioinformaticians or Biologists, you may be familiar with BLAST, a search algorithm that allows researchers to identify the group of organisms (species, taxa...) from which DNA/Protein sequences come.
๐ฅฑ You may also be familiar with the difficulties to interpret long and multi-parametric results coming out from BLAST searches: here's where we can operate with LLMs, summarizing the outputs and/or replying to queries about them!
๐งฌ You can now run BLAST for 16S rRNA bacterial sequences here on HF, summarizing and/or asking questions about the results, or make sense of your online BLAST searches uploading description tables, using the last space I built: as-cle-bert/BLAST-SummarAIzer
Have fun and may this be helpful to your research!๐ป
๐งฌ I'm thrilled to announce ๐ฝ๐ช๐ก๐ ๐๐ง๐ค๐ฉ๐๐๐ฃ๐ซ๐๐ฏ, a new functionality which supports multiple structure predictions at once: you just need to upload a FASTA file with all the amino-acidic sequences, and you'll be done in minutes!
๐ This can be really helpful in speeding up your research: give it a shot, if you are curious!๐ค
(Demo in the attached video)
๐ฌ I'm thrilled to introduce you: as-cle-bert/chat-with-em, the Space that lets you build a customizable chat model by specifying system instructions, temperature and maximum number of output tokens
๐ฆ๐ Thanks to LangChain, you can easily choose and switch among Claude models, Command-R, GPT-3.5, GPT-4o, Llama-3-8B, Llama-3-70B and Mixtral 8x7b: you just need to provide an API key!
Enjoy!๐ค
Hi!
I'm glad to hear your opinion: the idea behind this project is to provide help to everyone who seeks solutions to live a more climate-aware life, as well as to give advice to companies or investors who wanna focus on creating a greener future for their business.
I'm totally with you when you say this won't be the solution to the systemic problem of climate change (I will never claim that), but I feel like AI assistants like the one I built can really help us gather and organize knowledge and best practices to live and invest more sustainably, and build a better future for everyone and not just for ourselves as individuals. I really hope I was able to clarify the scopes of my project: let me know if you have any more questions!๐
๐ค ... And that's where AI comes into the play: we can indeed try to leverage, tweak and expand its knowledge in the field to extract valuable climate-aware solutions.
๐ค I tried to make something alike: exploiting climatebert/tcfd_recommendations as knowledge base, Qdrant Cloud as vector store service and microsoft/Phi-3-mini-128k-instruct as LLM (provided via API from eswardivi/Phi-3-mini-128k-instruct by @eswardivi ), I built an AI assistant to help you find climate-oriented solutions for your investments, companies, or simply for your everyday life๐.
Find it here: as-cle-bert/cLLiMateChat
GitHub: https://github.com/AstraBert/qdrant-ai-chat
Website: https://astrabert.github.io/qdrant-ai-chat/
Be kind to our Planet, we only got one๐
(Shout-outs to @JohnSmith9982 whose JohnSmith9982/small_and_pretty Gradio theme was used to build my application๐)
PS: ๐ฑCurious of knowing what is your carbon footprint? Head over to this ML-backed HF Space I built to discover it: as-cle-bert/carbon-footprint-predictor
๐ป If you are curious about the so-called "brainoware", a hardware built upon a brain organoid and used for AI and ML tasks, you may want to read my latest ๐ค article: https://huggingface.co/blog/as-cle-bert/brain-next-generation-neurons
๐กEnjoy!
What is everything-ai?
๐ค everything-ai is natively a multi-tasking agent, 100% local, that is able to perform several AI-related tasks
What's new?
๐ I am more than thrilled to introduce some new functionalities that were added since last release:
- ๐๏ธ๐ Handle audio files or microphone recordings, classifying or transcribing them with almost every audio-classification and automatic-speech-recognition model on Hugging Face Hub.
- ๐ฝ๏ธ Generate video from text prompts with almost every text-to-video model on HuggingFace Hub (original architecture by [Vasiliy Katsyka](https://github.com/Vasiliy-katsyka))
- ๐งฌ Predict the 3D structure of proteins from their amino-acidic sequence, with EsmFold by AI at Meta ([demo]( as-cle-bert/proteinviz)
- ๐๏ธ Finetune HF models on several downstream tasks with AutoTrain local integration (AutoTrain is developed by [Abhishek Thakur](https://github.com/abhishekkrthakur))
- ๐ฃ๏ธ Unleash powerful LLMs and exploit larger database collections for RAG with the integration of Hugging Face Spaces API and Supabase PostgreSQL databases ([demo](https://huggingface.co/spaces/as-cle-bert/supabase-ai-chat))
How can you use all of these features?
You just need a
docker compose up
!๐Where can I find everything I need?
Get the source code (and leave a little โญ while you're there):
https://github.com/AstraBert/everything-ai
Get a quick-start with the documentation:
https://astrabert.github.io/everything-ai/
Credits and inspiration
Shout-outs to Hugging Face, Gradio, Docker, AI at Meta, Abhishek Thakur, Qdrant, LangChain and Supabase for making all of this possible!
Inspired by: Jan, Cheshire Cat AI, LM Studio, Ollama and other awesome local AI solutions!
It fits perfectly, thank you soo much!๐ฅฐ๐ค (reference code: https://github.com/AstraBert/proteinviz/tree/v1.0.0)
That's really interesting! I'm gonna look it up and try it soon: thank you very much๐ฅฐ๐ค
I'm thrilled to share the latest updates regarding the Space I built for protein 3D structure prediction ( as-cle-bert/proteinviz): thanks to @lunarflu inputs, @osanseviero precious advice and @simonduerr 's article "Visualize proteins on Hugging Face Spaces" (https://huggingface.co/blog/spaces_3dmoljs, go check it out!), I was able to finally display the 3D protein models directly on-browser, without any need for fancy downloads of big HTMLs!
Take a look to the attached video, that shows how everything works, and make sure to visit the GitHub repository (https://github.com/AstraBert/proteinviz: leave a little โญ while you're there!)๐ฅฐ
May you have fun and luck with your protein research!๐งฌ
Okok, I knew about BRCA1 and its role in BC and, even though I'm not that much into cancer biology, this looks great and I'm really happy that my Space was of help to you๐ฅฐ
Stay tuned, as better-quality protein visualization is coming!๐
That's fantastic! Would be glad to hear more about your research!๐ค
So, just an update for @lunarflu and everyone else who is interested:
- Added three examples (encompassing human hemoglobin, mouse GTPase and a human ribosomal protein)
- Added more in depth example table on the GitHub repo
- Added support for 3D rendering: now you get to download an HTML file (approx. 3.5 MB) with the 3D model to explore and play with... Here is an example:
Hi @lunarflu , thanks for the feedback!
So, to reply to your questions:
- There are 20 amino-acids (represented by letters) that one can use + some rare or degenerate (meaning you are unsure what letter to put) ones. You can find them here: https://bredagenetics.com/amino-acid-codes-symbols/
- I did not think of that, but I'll sure do :)) Thank you very much for the suggestion!
In the meanwhile, if you are interested in some example proteins to try out, I suggest using Uniprot and searching for some proteins like "Insulin", "Beta-lactamase" or "Hemoglobin". There should be the amino-acidic sequence and the actual 3D structure that you can compare!
If you are excited about AlphaFold3, but upset because it is not open-source, I might have a solution to cheer you up a little bit:
as-cle-bert/proteinviz
This is a space that lets you predict the 3D structure of proteins from their amino-acidic sequences, with the protein folding model facebook/esmfold_v1: using this space is the perfect quick-start to become a Protein Scientist! (or maybe not, who knows...๐ค)
In the meantime, if you are curious about what's going on with AlphaFold3 and want something Biologist๐ฌ/Computer Scientist๐ป-friendly, you can also check out the latest community blog post I wrote: https://huggingface.co/blog/as-cle-bert/what-is-going-on-with-alphafold3 ๐
Have fun and enjoy open-source science!๐งฌ
Yes, definitely!
These days I'm compiling a collection of spaces to showcase the most important tasks! Feel free to check it out at: https://huggingface.co/collections/as-cle-bert/everything-ai-66465ad7e555bad40d7bccf1 ๐ค
Did you know that you are a
docker compose up
away from unleashing some really powerful AI?Let me introduce you to ๐ฒ๐๐ฒ๐ฟ๐๐๐ต๐ถ๐ป๐ด-๐ฎ๐ถ๐ค
๐ฒ๐๐ฒ๐ฟ๐๐๐ต๐ถ๐ป๐ด-๐ฎ๐ถ is a multi-task, AI-powered and local assistant that allows you to:
- ๐ญ talk to ๐บ๐ฐ๐ถ๐ณ ๐ฑ๐ฅ๐ง๐ด thanks to a backend retrieval architecture built with LangChain and Qdrant (supports 50+ languages)
- โ๏ธ chat with almost every text-generation model on Hugging Face hub (supports 50+ languages)
- ๐ summarize texts and pdfs with almost every text-summarization model on HF hub (English-only)
- ๐จ generate images with Pollinations AI API or stable diffusion models on HF hub (supports 50+ languages)
- ๐จ๏ธ describe images with image-to-text models on HF hub (English-only)
- ๐ฉป classify images with image-classification models on HF hub (English-only)
- ๐ผ๏ธ implement an image-based search engine on *your images*, thanks to Qdrant, with almost every feature extraction model on HF hub
๐ฅ๏ธ There is now a brand new user interface that helps with selecting tasks, models, pdf files, images and languages
๐ everything-ai is distributed as a multi-container application in Docker, available for Windows and Mac, and it is 100% local.
๐โโ๏ธ Head over to the documentation for a quick-start on the application: https://astrabert.github.io/everything-ai
๐ฅฐ Make sure to check out the GitHub repo and leave a little โญ: https://github.com/AstraBert/everything-ai
๐ Features coming soon:
- ๐๏ธAudio and video-related tasks
- ๐ค Integration with HF Spaces API to effortlessly chat with larger models on the Hub from your local computer
- ๐งฌ Biology-related tasks
โค๏ธ Shout-outs to Qdrant, LangChain, Hugging Face, Docker and Gradio for making all of this possible!
๐กInspired by Jan, Cheshire Cat AI, LM Studio, Ollama and other awesome local AI solutions!