Unverified50% confidenceBenchmarkExact time
该量化模型相较于BF16全精度版本保留了约77%的准确率
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/24/2026
First Seen
Valid until: 11/22/2026
Sources
Related Entities
Related Claims
Verified模型参数量越大，对量化的鲁棒性越强，使得70B级别模型在4-bit量化后仍能保持接近FP16的性能72% similarUnverifiedNVFP4量化方案相比BF16基线实现了1.5倍的推理加速，同时在top-1准确率上保留了92%到97%的水平70% similarUnverifiedtop-1%准确率是评估量化模型与原始全精度模型输出一致性的指标，关注模型置信度最高的1%的token是否与全精度版本一致67% similarUnverified在同一模型（Claude 4.6 Opus）下，OCR的精度（Precision）为33.9%，通用Agent为7.23%，OCR精度提升约4.7倍65% similarUnverifiedModels like GPT-4 and Claude 3.5 surpassed 90% accuracy on MMLU, significantly diminishing its discriminative power64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/795615API
curl https://kongchang.com/api/v1/knowledge/claims/795615MCP
get_claim(id=795615)