Chang Xu

Add Quant Row&Column ParallelLinear

1

### PR types New features ### PR changes APIs ### Describe add QuantRowParallelLinear and QuantColumnParallelLinear in Quantization; support model parallel QAT

Quant Analyzer provides: 1. compute the sensitivity of the model by quantizing layer by layer; (sensitivity is expressed as model accuracy for now) 2. rank the sensitivity which can show...

add picodet-s-npu full quant demo

1. 手动设置搜索空间(允许block wise等方式搜索) 2. 自动tokenize子网络 3. 支持根据输入token跑特定子网 4. BN和BN2D允许在同一网络结构中，能够转换为对应super op（Det中backbone常使用BN，neck head常使用BN2D） 5. 支持不同layer/block可以搜索不同搜索空间 6. export模型支持group裁剪