Lossless LLM compression for efficient GPU inference via dynamic-length float (arxiv.org)

The AI summary for this story hasn't been generated yet — it's produced hourly. Check back soon. Meanwhile, read the discussion on HN.