Popular Open Source Models
Master Qwen 2.5, DeepSeek-R1, GLM-4, Gemma 2, and Llama 3 — sweet spots, architectures, and local hardware requirements
4 modules8 lessons10 questions
01 The Qwen Ecosystem (Alibaba)
Qwen 2.5 & Qwen 2.5-Coder
2 qQwen 2.5 from Alibaba is currently the reigning open-source coding champion! Its 32B Coder model can write entire apps, debug complex code, and fits on a single high-end GPU or Mac. 💻🏆
Qwen 2.5 Math & 1M Long-Context
1 qQwen 2.5 can read entire book series or codebases in one go thanks to its 1-million-token context window, while its Math model solves PhD-level equations step-by-step! 📚📐
02 The DeepSeek Breakthroughs
DeepSeek-R1: Open Reasoning Architecture
2 qDeepSeek-R1 is the open-source reasoning model that thinks before answering. It shows its inner thought process with <think> tags, catching its own mistakes just like OpenAI's o1! 🧠🔍
DeepSeek-V3 & DeepSeek-Coder-V2 (MoE)
1 qDeepSeek-V3 is a giant 671B model that only activates 37B weights for each word. That means you get the intelligence of a massive brain at the running speed of a small one! ⚡
03 Gemma & GLM Families
Google Gemma 2 (2B, 9B, 27B)
1 qGemma 2 from Google DeepMind punches way above its weight. The 27B model outperforms much larger models while fitting completely onto a single 24GB graphics card! 💎
GLM-4 & GLM-4V (Zhipu AI)
1 qGLM-4 is a multilingual powerhouse that excels at understanding photos, charts, and complex tools with precise accuracy right on your laptop! 🖼️🛠️
04 Meta Llama & Mistral Foundations
Llama 3.1 & 3.3 (8B, 70B)
1 qMeta's Llama 3 series is the universal baseline of open-source AI. Almost every local app, tool, and fine-tuning tutorial is built to support Llama 3 first! 🦙
Mistral NeMo & Mixtral MoE
1 qMistral from France built the pioneer Mixture-of-Experts models. Their 12B NeMo and Mixtral 8x7B models deliver extreme efficiency with their high-speed Tekken tokenizer! 🥐⚡