#from-scratch
GitHub repositories that have self-applied the topic "from-scratch" — a creator-tagged metadata that surfaces how AI projects describe themselves.
REPOS Repos for #from-scratch (top 10 by stars)
A 1.32B-parameter Mixture-of-Experts LLM trained from scratch in PyTorch - custom tokenizer, MLA attention, 18B-token pretraining, SFT and DPO. No pretrained weights, no HuggingFace Trainer.
atandra2000/GPT-OSS-LiteFaithful from-scratch PyTorch reproduction of OpenAI's GPT-OSS architecture (sliding/full attention alternation, learned attention sinks, YaRN 128K, top-2-of-8 MoE), scaled to Chinchilla-optimal 502M total / 247M active training on a single A100 80GB
monkfromearth/monkwayMonk LLM Gateway: one API in front of every model. An LLM gateway and router built from scratch (proxy, routing, retry/fallback, caching, budgets), with an interactive course.
aurimas13/Calculator-AgentCalculator Agent: First AI agent, built from scratch. Claude using a calculator tool in a loop, with every step streamed live to a web UI. No agent framework.
zch6a/build-your-own-aiBuild your own AI from scratch: tokenizer, autograd, GPT, inference, vector database, RAG, agent and MCP. Each a small, tested, readable module that runs on a laptop CPU with no API key.
KeigoShimadaCC/llm-from-scratchMac-local LLM-from-scratch lab for building and evaluating a small decoder-only language model on Apple Silicon.
iqureshi123/tinyinfer#1335284915An LLM inference engine written from scratch in Python and C++/Metal — tokenizer, forward pass, KV cache, and INT4 quantization implemented by hand. No PyTorch, no llama.cpp.
iqureshi123/tinyinferAn LLM inference engine written from scratch in Python and C++/Metal — tokenizer, forward pass, KV cache, and INT4 quantization implemented by hand. No PyTorch, no llama.cpp.
PrayasPanda/tinytalkA 14M-parameter GPT built from scratch in PyTorch hand-written attention, custom BPE tokenizer, no pretrained weights.
DevxD98/JimmyLabsJimmyLabs — a GPT-style language model built from scratch and fully documented on a MacBook Air M1 (8GB, Apple Silicon). Documentation-first educational research platform.
RELATED Other topics · full topics ranking →
#claude-code
3,787#ai-agents
3,201#llm
2,090#mcp
1,866#claude
1,854#python
1,678#developer-tools
1,530#codex
1,376#ai
1,181Aggregated by case-insensitive match against topics_json of each repo's latest content snapshot. methodology