Unverified50% confidenceFactExact time
AWQ（Activation-aware Weight Quantization）于2023年提出，发现权重中约 1% 的显著通道对精度影响最大
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedAWQ由MIT团队提出，发现仅约1%的权重对激活值影响显著，通过保护这部分关键权重实现更优的低比特量化效果69% similarUnverifiedThe core insight of AWQ is that only approximately 1% of salient weights have a major impact on model output, and protecting them enables model quality to be maintained at extremely low bit widths66% similarUnverifiedThe L1 vector router achieves an accuracy of over 95% on the top-20 high-frequency intents, only 1.5 percentage points below the L2 large model, but with approximately 30× lower latency and cost.55% similarUnverified参数量在1亿以下的轻量模型经领域数据微调后可达95%以上意图分类准确率，单次推理延迟控制在5-20ms以内55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566751API
curl https://kongchang.com/api/v1/knowledge/claims/566751MCP
get_claim(id=566751)