1
Show HN: We matched full-context recall on ~1% of the tokens (open benchmark)
用1%的token复现全上下文回忆，开源基准实测有效，直击长上下文“lost in the middle”痛点。
Article URL: https://github.com/compresh/compresh-benchmarks/blob/main/epbench/WRITEUP.md Comments URL: https://news.ycombinator.com/item?id=48641509 …