search

intel-analytics / ipex-llm

Accelerate local LLM inference and finetuning (LLaMA, Mistral, ChatGLM, Qwen, Mixtral, Gemma, Phi, MiniCPM, Qwen-VL, MiniCPM-V, etc.) on Intel XPU (e.g., local PC with iGPU and NPU, discrete GPU such as Arc, Flex and Max); seamlessly integrate with llama.cpp, Ollama, HuggingFace, LangChain, LlamaIndex, vLLM, GraphRAG, DeepSpeed, Axolotl, etc
Apache License 2.0
6.71k stars 1.26k forks source link

There is no federated learning in the PPML test case of version 2.2.0. #8893

Open zhengjujing opened 1 year ago

zhengjujing commented 1 year ago

There is no federated learning in the PPML test case of version 2.2.0. May I ask if federated learning is not supported in PPML?

qiyuangong commented 1 year ago

Hi @zhengjujing . Thank you for your feedback. :)

Federated learning is the main building block of PPML. Most code and design are public at https://github.com/intel-analytics/BigDL/tree/main/python/ppml/src/bigdl/ppml .

Federated Learning test cases are in these locations:

zhengjujing commented 1 year ago

Thank you for your support. I have seen that federated learning includes some algorithms for machine learning and deep learning. During the federated modeling process, communication between all parties is established through the FLServer class, which is a GRPC channel protected by tls. What I would like to ask is whether EHSM (kms and as) authentication process is required when using already supported federated modeling algorithms?

qiyuangong commented 1 year ago

Thank you for your support. I have seen that federated learning includes some algorithms for machine learning and deep learning. During the federated modeling process, communication between all parties is established through the FLServer class, which is a GRPC channel protected by tls. What I would like to ask is whether EHSM (kms and as) authentication process is required when using already supported federated modeling algorithms?

Hi @zhengjujing

Authentication is required for Federated Learning to avoid unauthorized parties participating in the training process and accessing sensitive information. However, there are multiple solutions for Authentication, e.g., mTLS, Authentication servers, etc. Choosing proper authentication solutions depends on existing ecosystems, such as existing authentication servers or certificates.

Without losing generality, we prepare an FL example (https://bigdl.readthedocs.io/en/latest/doc/PPML/Overview/trusted_fl.html) based on TLS. Customers can enable mTLS if they want. They can further upgrade this example using EHSM (KMS and attestation service running in SGX) or their own service.