Unverified50% confidenceBenchmarkExact time
On LongMemEval, CTWM achieved a 24.48% token reduction while maintaining overall accuracy at the same level.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the LongMemEval benchmark, BM25 and embedding solutions converge to approximately 95% accuracy at 20 chunks67% similarUnverifiedAfter applying logit penalties, the number of reasoning tokens decreased by 11% to 19% across all precision formats.63% similarUnverified传统 LLM 推理中因需按最大长度预留显存，KV Cache 的实际利用率通常不到 30%61% similarUnverifiedGGUF quantization can reduce a model's memory footprint to approximately 25% of its original precision.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966675API
curl https://kongchang.com/api/v1/knowledge/claims/966675MCP
get_claim(id=966675)