Today's LLM & Language Models: Fastest-Growing Projects — April 21, 2026
Today's the LLM & Language Models space, we're seeing a surge in interest around personal knowledge bases and tools that enable more natural human-like conversations. Many of the fastest-growing repositories are focused on helping users compile and organize their own knowledge, while others aim to make AI-generated text sound more authentic.
One standout repository is arman-bd/guppylm, which boasts an impressive Growth Score of 83.80 and has garnered 2,967 stars. This ~9M parameter LLM talks like a small fish, offering a unique approach to language generation that's clearly resonating with developers. With 24 commits in the past 30 days, it's clear that this project is actively being improved and expanded.
Another rapidly growing repository is hexiecs/talk-normal, which has achieved a Growth Score of 70.08 and attracted 1,372 stars. This tool makes any LLM talk like a normal person by removing AI-generated slop, making it an attractive solution for those looking to create more human-like interactions. With 56 commits in the past month, this project is clearly being actively developed.
sdyckjq-lab/llm-wiki-skill has also seen significant growth, with a Growth Score of 58.09 and 950 stars. This personal knowledge base skill supports multiple platforms and leverages Karpathy's llm-wiki methodology, offering users a powerful tool for organizing their own knowledge.
Meanwhile, amitshekhariitbhu/llm-internals has earned a Growth Score of 44.56 and 601 stars by providing an in-depth look at LLM internals, from tokenization to attention to inference optimization. This repository offers developers a valuable resource for understanding the underlying mechanics of language models.
Other notable repositories include kessler/gemma-gem, which runs Google's Gemma 4 model entirely on-device via WebGPU (Growth Score: 37.62, Stars: 811); xoai/sage-wiki, an LLM-compiled personal knowledge base that extracts concepts and discovers cross-references (Growth Score: 37.21, Stars: 443); and VectifyAI/OpenKB, an open LLM knowledge base with a Growth Score of 33.18 and 456 stars.
Rounding out the top 10 are lucasastorian/llmwiki, an open-source implementation of Karpathy's LLM Wiki (Growth Score: 31.68, Stars: 585); atomicmemory/llm-wiki-compiler, a knowledge compiler that generates interlinked wikis from raw sources (Growth Score: 30.72, Stars: 650); and Pratiyush/llm-wiki, an LLM-powered knowledge base built on Claude Code and other platforms (Growth Score: 30.19, Stars: 134).
One standout repository is arman-bd/guppylm, which boasts an impressive Growth Score of 83.80 and has garnered 2,967 stars. This ~9M parameter LLM talks like a small fish, offering a unique approach to language generation that's clearly resonating with developers. With 24 commits in the past 30 days, it's clear that this project is actively being improved and expanded.
Another rapidly growing repository is hexiecs/talk-normal, which has achieved a Growth Score of 70.08 and attracted 1,372 stars. This tool makes any LLM talk like a normal person by removing AI-generated slop, making it an attractive solution for those looking to create more human-like interactions. With 56 commits in the past month, this project is clearly being actively developed.
sdyckjq-lab/llm-wiki-skill has also seen significant growth, with a Growth Score of 58.09 and 950 stars. This personal knowledge base skill supports multiple platforms and leverages Karpathy's llm-wiki methodology, offering users a powerful tool for organizing their own knowledge.
Meanwhile, amitshekhariitbhu/llm-internals has earned a Growth Score of 44.56 and 601 stars by providing an in-depth look at LLM internals, from tokenization to attention to inference optimization. This repository offers developers a valuable resource for understanding the underlying mechanics of language models.
Other notable repositories include kessler/gemma-gem, which runs Google's Gemma 4 model entirely on-device via WebGPU (Growth Score: 37.62, Stars: 811); xoai/sage-wiki, an LLM-compiled personal knowledge base that extracts concepts and discovers cross-references (Growth Score: 37.21, Stars: 443); and VectifyAI/OpenKB, an open LLM knowledge base with a Growth Score of 33.18 and 456 stars.
Rounding out the top 10 are lucasastorian/llmwiki, an open-source implementation of Karpathy's LLM Wiki (Growth Score: 31.68, Stars: 585); atomicmemory/llm-wiki-compiler, a knowledge compiler that generates interlinked wikis from raw sources (Growth Score: 30.72, Stars: 650); and Pratiyush/llm-wiki, an LLM-powered knowledge base built on Claude Code and other platforms (Growth Score: 30.19, Stars: 134).