From f19f3e71a9bd92afc3f3978881180b587f3bf36c Mon Sep 17 00:00:00 2001
From: Xiao-Chenguang <973990275@qq.com>
Date: Tue, 22 Oct 2024 17:56:57 +0100
Subject: [PATCH] add instruction to readme

---
 README.md | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index d6dee6f..bc5a472 100644
--- a/README.md
+++ b/README.md
@@ -23,6 +23,28 @@ Make a copy of both the [config.yaml](./config.yaml) and [fedavg_demo.py](./exam
 python fedavg_demo.py
 ```
 
+Here we recommend you to use the [UV](https://docs.astral.sh/uv/) as a python environment manager to create a clean environment for the experiments.
+
+After install `uv`, you can create a new environment and run a `FedMind` example with the following command:
+```bash
+uv init FL-demo
+cd FL-demo
+source .uv/bin/activate
+
+uv add fedmind torchvision
+wget https://raw.githubusercontent.com/Xiao-Chenguang/FedMind/refs/heads/main/examples/fedavg_demo.py
+wget https://raw.githubusercontent.com/Xiao-Chenguang/FedMind/refs/heads/main/config.yaml
+
+uv run python fedavg_demo.py
+```
+
+
 ## Features
-- Parallel training speed up.
-- Serialization for limited resources.
\ No newline at end of file
+This FL framework provides two client simulation modes depending on your resources:
+- Parallel training speed up for powerful resources.
+- Serialization for limited resources.
+
+This is controlled by the parameter `NUM_PROCESS` which can be set in the [config.yaml](./config.yaml).
+Setting `NUM_PROCESS` to 0 will use the serialization mode where each client trains sequentially in same global round.
+Setting `NUM_PROCESS > 0` will use the parallel mode where `NUM_PROCESS` workers consume the clients tasks in parallel.
+The recommended value for `NUM_PROCESS` is the number of **CPU cores** available.