Haodong Zhao

FATE-LLM-pellm

3

求问，在FATE-LLM中的pellm中，将base_model替换为添加了adapter（如lora等）的peft_model，在客户端训练后、聚合时仅传递给aggregator adapter部分的参数、不传递base_model参数是在哪里实现的？

Add tutorial Split Learning for bank marketing-torch

11

Add tutorial Split Learning for bank marketing-torch # Preamble For Pull Request initiator, please read the following steps carefully. 请合并请求的发起人仔细阅读下列步骤。 --- If you are creating a `Pull Request` about listed...

【OSCP】拆分学习 SPlitGuard 检测防御算法实现

9

Fixed #776 在secretflow中实现拆分学习 SPlitGuard 检测防御算法 # Preamble For Pull Request initiator, please read the following steps carefully. 请合并请求的发起人仔细阅读下列步骤。 --- If you are creating a `Pull Request` about listed topic below,...

SLModel的torch backend是否支持Naive SplitNN架构

4

### Issue Type Feature Request ### Source binary ### Secretflow Version 1.2.0.dev20230918 ### OS Platform and Distribution Centos7 ### Python version 3.8.18 ### Bazel version _No response_ ### GCC/Compiler version...

secretflow/ml/nn/sl/backend/torch

1

### Issue Type Feature Request ### Source source ### Secretflow Version 0.8.1b0 ### OS Platform and Distribution Centos7 ### Python version 3.8.13 ### Bazel version _No response_ ### GCC/Compiler version...

按照教程https://zhuanlan.zhihu.com/p/621793987 使用lora对ChatGLM微调后，模型的answer全部为很多个 ?? 

Does peft supports the custom setting of trainable parameters(for example, some params in word_embeddings)

2

### Feature request Does peft supports the custom setting of trainable parameters(for example, some params in word_embeddings) ### Motivation To use the method [EP](https://arxiv.org/abs/2103.15543) ### Your contribution maybe

Add research report on horizontal federated learning backdoor attack and defense. @zhaocaibei123