KV Cache Compression 900000x Beyond TurboQuant and Per-Vector Shannon Limit

Article

• Claims KV cache compression exceeding 900,000x over prior state-of-the-art
• Uses the model itself as the compression dictionary
• Purports to exceed the per-vector Shannon entropy limit

Discussion

• Commenters intrigued; argument hard to follow
• Model-as-dictionary is the novel framing
• Skepticism about extraordinary claims without clear proof

Discuss on HN