Practical offloading for fine-tuning LLM on commodity GPU via learned sparse projectors

Re-assign community