At 1 bpd on LLaMA-70B, Qwen-72B, and Mixtral-8x7B, KV-COBRA variants remained much closer to FP16 than the collapsing integer-grid baselines on perplexity, and KV-COBRA_KL was the best compressed method for PPL and LongBench in every family.

At 1 bpd on LLaMA-70B, Qwen-72B, and Mixtral-8x7B, KV-COBRA variants remained much closer to FP16 than the collapsing integer-grid baselines on perplexity, and KV-COBRA_KL was the best compressed method for PPL and LongBench in every family. · CiteArk

测量结果

已评估 0/68 项

获得支持 0有争议或冲突 0尚未评估 68

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
bpd: 16 · side: K-only · model: LLaMA-3.1-70B · method: FP16	5.38 perplexity	—	尚未评估
bpd: 16 · side: K-only · model: LLaMA-3.1-70B · method: FP16	60.4%	—	尚未评估
bpd: 16 · side: K-only · model: LLaMA-3.1-70B · method: FP16	14.3 F1	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_KL	6.97 perplexity	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_KL	58.4%	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_KL	13.5 F1	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_MSE	9.5 perplexity	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_MSE	58.5%	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KV-COBRA_MSE	10.5 F1	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KQ-SVD	9.79 perplexity	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KQ-SVD	57%	—	尚未评估
bpd: 1 · side: K-only · model: LLaMA-3.1-70B · method: KQ-SVD	11.2 F1	—	尚未评估

At 1 bpd on LLaMA-70B, Qwen-72B, and Mixtral-8x7B, KV-COBRA variants remained much closer to FP16 than the collapsing integer-grid baselines on perplexity, and KV-COBRA_KL was the best compressed method for PPL and LongBench in every family.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果