Unverified50% confidenceFactExact time
The impact of compression on a model's internal knowledge distribution is non-uniform: some capabilities are almost lossless, others may degrade substantially, while the average score may show only small fluctuations.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
Unverified小模型在分布内高性能的同时存在分布外脆弱性，一旦推理时的工具描述措辞、参数命名或用户查询风格偏离训练集，成功率可能显著下滑74% similarUnverifiedQuantization errors are usually distributed uniformly across all parameters, whereas the destructiveness of pruning is more concentrated on the removed connections, so the two have different patterns of impact on knowledge distribution.74% similarUnverifiedWhen information is insufficient, the content generated by large models tends to go off track; the poorer the context, the worse the output quality.74% similarUnverifiedWhen the context approaches its limit, the model may forget critical early information or produce lower-quality output.73% similarUnverified模型压缩技术（量化、剪枝、知识蒸馏）会不成比例地损害模型的事实准确性，量化后模型在事实性问答任务上的幻觉率可能显著上升73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1079505API
curl https://kongchang.com/api/v1/knowledge/claims/1079505MCP
get_claim(id=1079505)