fix cli distributed launch without rank info (#3714)

Author: Jonathans575

docs/en/cli_usage.md

@@ -107,14 +107,14 @@ paddleformers-cli export examples/config/run_export.yaml
 #### 6.1. Method 1
 
 ```bash
-NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
+NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} RANK={rank} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
 ```
 
 #### 6.2. Method 2 (mpirun)
 
 First, write a script, such as `scripts/train_96_gpus.sh`, with the following content:
 ```bash
-NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
+NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} RANK={rank} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
 ```
 
 Then:

docs/zh/cli_usage.md

@@ -92,15 +92,15 @@ paddleformers-cli export examples/config/run_export.yaml
 #### 方式一
 
 ```shell
-NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
+NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} RANK={rank} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
 ```
 
 #### 方式二 (mpirun)
 
 先写一个脚本，例如`scripts/train_96_gpus.sh`，内容为：
 
 ```shell
-NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
+NNODES={num_nodes} MASTER_ADDR={your_master_addr} MASTER_PORT={your_master_port} RANK={rank} CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 paddleformers-cli train examples/config/sft_full.yaml
 ```
 
 然后：

paddleformers/cli/cli.py

@@ -71,6 +71,7 @@ def main():
     distributed_funcs = ["train", "export"]
     paddleformers_dist_log = os.getenv("PADDLEFORMERS_DIST_LOG", "paddleformers_dist_log")
     nnodes = os.getenv("NNODES", "1")
+    rank = os.getenv("RANK", "0")
     master_ip = os.getenv("MASTER_ADDR", "127.0.0.1")
     master_port = os.getenv("MASTER_PORT", "8080")
     current_device = detect_device()
@@ -154,7 +155,7 @@ def main():
         command = (
             f"python -m paddle.distributed.launch --log_dir {paddleformers_dist_log} "
             f"--{current_device}s {visible_cards} --master {master_ip}:{master_port} "
-            f"--nnodes {nnodes} {launcher.__file__} {args_to_pass}"
+            f"--nnodes {nnodes} --rank {rank} --run_mode=collective {launcher.__file__} {args_to_pass}"
         )
         command = shlex.split(command)
         process = subprocess.Popen(