@WhaleBarnacle@carrigmat@InAllShadows@peepee_p00p00_@mgoetzke Just tried the online official version of R1. It took a little over 10k tokens (~270 seconds) in the "MoE thinking" procees to write a 30-40 lines python function for me. Assuming 10 tokens per second if hosted locally, to get the same response, it will spend ~17 mins thinking.
Complete hardware + software setup for running Deepseek-R1 locally. The actual model, no distillations, and Q8 quantization for full quality. Total cost, $6,000. All download and part links below: