> "Complete hardware + software setup for running Deepseek-R1 locally. The actual model, no distillations, and Q8 quantization for full quality. Total cost, $6,

> "Complete hardware + software setup for running Deepseek-R1 locally. The actual model, no distillations, and Q8 quantization for full quality. Total cost, $6,000." These guys say they got really good results with 2.51-bit quantization of the original R1. The original has 671B params weighing in at 720GB - that's what they're running on this $6000 setup. According to these guys[0] they get really good results at 2.51bit quantization which would be this model[1] which is still 671B params, but weighs in at 212GB. [0] https://unsloth.ai/blog/deepseekr1-dynamic https://unsloth.ai/blog/deepseekr1-dynamic [1] https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/DeepSeek-R1-UD-Q2_K_XL https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/De...