CensorAI is a filter model created and trained on 1,000 examples, exclusively to recognize patterns that pose a real danger to the user. It predicts the likelihood of such harm in text. CensorAI is a secure system: it doesn’t rely on static dictionaries or hardcoded phrases, but operates based on trained neural network parameters. These are full-fledged text analysis models built on Nexora, with an architecture based on embedding and dense layers, containing over 100,000 active parameters. CensorAI is a project containing several language models of varying power, accuracy, and quality, each model running locally on the user's device. #Nexora #AI #CensorAI #FromScratch #FamilyAI #100K+Parameters #ThoughtfulDesign #ResponsiveChat #UIDesign
Models from CensorAI: - CensorAI:141K is a powerful, advanced filtering model with 140,993 trainable parameters and six layers, featuring Embedding, Dense, and BatchNorm components. It processes text within a 25-character context window. - CensorAI:112K is a lightweight alternative with 112,417 trainable parameters and four layers (Embedding and Dense only), also using a 25-character context window.
