Results of the project briefly described. Conversion of a scanned text to structured data ( #Wikidata ), using #HTR ( #eScriptorim ) and #text2text #transformers

Funded by #Wikimedia.

#CzechAcademyOfSciences
https://meta.wikimedia.org/wiki/Research:Wikimedia_versus_traditional_biographical_encyclopedias

There's an ongoing drive for #FOSS #distributedAI computing, e.g. #aiHorde. It's not perfect & needs a 3rd-party security audit, but is somewhere on the path to more-secure non-local #edgecomputing for devices like phones that can't load large #AImodels or run popular #text2text / #text2image #genAI applications locally.

*NB: Apologies to Horde volunteers if I just bumped your traffic past reasonable limits.*

https://aihorde.net

There's an ongoing drive to create #FOSS open source, #distibutedAI computing. #AI_Horde is one such project, and possibly the most well-known right now. It's not perfect, and no one I know has done an in-depth security analysis, but I suspect this is somewhere on the path to safer non-local AI computing for edge devices like phones that simply can't run larger #AImodels or popular #text2text or #text2image #genAI applications locally.

NB: My apologies to the Horde volunteers if I just bumped your traffic past reasonable limits.

https://aihorde.net

✨ Open source RAG (Retrieval Augmented Generation) right in your browser! ✨

#SemanticFinder now offers an 𝐚𝐝𝐯𝐚𝐧𝐜𝐞𝐝 𝐜𝐡𝐚𝐭 & 𝐬𝐮𝐦𝐦𝐚𝐫𝐲 𝐟𝐞𝐚𝐭𝐮𝐫𝐞 for your search results - all in your browser.

💡There are very few capable small LLMs that offer high-quality results. Quantized LaMini-Flan-T5-783M offers good performance with 3-4s load time and >6 tokens/s after model download on an old i7.

https://do-me.github.io/SemanticFinder/

#transformers #RAG #AI #LLM #embeddings #semanticsearch #text2text #Flan #T5