update readme

doc/en/balance-serve.md

@@ -100,8 +100,10 @@ git submodule update --init --recursive
 
 # Install single NUMA dependencies
 USE_BALANCE_SERVE=1  bash ./install.sh
+pip install third_party/custom_flashinfer/
 # For those who have two cpu and 1T RAM（Dual NUMA）:
 USE_BALANCE_SERVE=1 USE_NUMA=1 bash ./install.sh
+pip install third_party/custom_flashinfer/
 ```
 
 ## Running DeepSeek-R1-Q4KM Models

doc/en/install.md

@@ -117,11 +117,13 @@ Download source code and compile:
 
     ```shell
     USE_BALANCE_SERVE=1 bash ./install.sh
+    pip install third_party/custom_flashinfer/
     ```
   - For Multi-concurrency with two cpu and 1T RAM:
 
     ```shell
     USE_BALANCE_SERVE=1 USE_NUMA=1 bash ./install.sh
+    pip install third_party/custom_flashinfer/
     ```
 - For Windows (Windows native temporarily deprecated, please try WSL)
 

doc/en/llama4.md

@@ -67,9 +67,11 @@ pip3 install torch torchvision torchaudio --index-url https://download.pytorch.o
 
 ```bash
 # Install single NUMA dependencies
-USE_BALANCE_SERVE=1  bash ./install.sh
+USE_BALANCE_SERVE=1  bash ./
+pip install third_party/custom_flashinfer/
 # For those who have two cpu and 1T RAM（Dual NUMA）:
 USE_BALANCE_SERVE=1 USE_NUMA=1 bash ./install.sh
+pip install third_party/custom_flashinfer/
 ```
 
 ### 4. Use our custom config.json

doc/zh/DeepseekR1_V3_tutorial_zh.md

@@ -127,8 +127,10 @@ cd ktransformers
 git submodule update --init --recursive
 # 如果使用双 numa 版本
 USE_BALANCE_SERVE=1 USE_NUMA=1 bash ./install.sh
+pip install third_party/custom_flashinfer/
 # 如果使用单 numa 版本
 USE_BALANCE_SERVE=1 bash ./install.sh
+pip install third_party/custom_flashinfer/
 # 启动命令
 python ktransformers/server/main.py --model_path <your model path> --gguf_path <your gguf path> --cpu_infer 62 --optimize_config_path <inject rule path> --port 10002 --chunk_size 256 --max_new_tokens 1024 --max_batch_size 4 --port 10002 --cache_lens 32768 --backend_type balance_serve
 ```