Comment by acb12

What are the minimal hardware requirements to run a reasonable good local model for real SWE work these days? And have a reasonable inference speed? Seems like the requirement are pretty high and the inference speed is not ideal.