typeof

Engineer on Base 🔵 • Building BlockSocks 🧦 • Denver, CO 📍

what's the biggest r1 that comfortably works? R1 70b crashes Zed on my 64GB

Farcaster Development Boutique @ https://dTech.vision 💜 in love with tech 🤓decentralization, fitness and oldschool rap 😎

I've got the DeepSeek-R1-Distill-Llama-70B-GGUF with Q4_K_M quantization loaded right now. It's generating ~10 tokens/sec. RAM usage is at 57%

what are you doing with the model? so far I find it fun to play with but have no direct applied wow moment

I haven’t done anything with the DeepSeek model so far but I’ve been using the Llama models to learn to build agents. Using RAG and stuff like that

The goal is to make a local Devin that can improve my codebases while I sleep