From 647525952d78a1878d82d6c835be0541d1a2e37e Mon Sep 17 00:00:00 2001
From: wql <wengqinlan18@163.com>
Date: Thu, 19 Sep 2024 15:25:12 +0800
Subject: [PATCH] chore: set ASCEND_RT_VISIBLE_DEVICES

---
 batch_run.sh | 22 ++++++++++++++++++++++
 run_once.sh  |  4 ++--
 2 files changed, 24 insertions(+), 2 deletions(-)

diff --git a/batch_run.sh b/batch_run.sh
index 56c43e92..b49ed766 100644
--- a/batch_run.sh
+++ b/batch_run.sh
@@ -4,6 +4,8 @@
 
 repeat_cnt=$1
 
+export ASCEND_RT_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+
 for ((i=0; i<repeat_cnt; i++)); do
     bash run_once.sh lora_sft Qwen-7B 8 500
     bash run_once.sh lora_sft Qwen-7B 8 1000
@@ -22,4 +24,24 @@ for ((i=0; i<repeat_cnt; i++)); do
     bash run_once.sh inference Baichuan2-7B 8 1000
     bash run_once.sh inference Baichuan2-7B 1 500
     bash run_once.sh inference Baichuan2-7B 1 1000
+
+    bash run_once.sh lora_sft chatglm2-6B 8 500
+    bash run_once.sh lora_sft chatglm2-6B 8 1000
+    bash run_once.sh lora_sft chatglm2-6B 1 500
+    bash run_once.sh lora_sft chatglm2-6B 1 1000
+    bash run_once.sh lora_sft llama-2-7b-ms 8 500
+    bash run_once.sh lora_sft llama-2-7b-ms 8 1000
+    bash run_once.sh lora_sft llama-2-7b-ms 1 500
+    bash run_once.sh lora_sft llama-2-7b-ms 1 1000
+
+    bash run_once.sh inference chatglm2-6B 8 500
+    bash run_once.sh inference chatglm2-6B 8 1000
+    bash run_once.sh inference chatglm2-6B 1 500
+    bash run_once.sh inference chatglm2-6B 1 1000
+
+    bash run_once.sh inference llama-2-7b-ms 8 500
+    bash run_once.sh inference llama-2-7b-ms 8 1000
+    bash run_once.sh inference llama-2-7b-ms 1 500
+    bash run_once.sh inference llama-2-7b-ms 1 1000
+
 done
diff --git a/run_once.sh b/run_once.sh
index e859f6e5..7a5bf72c 100644
--- a/run_once.sh
+++ b/run_once.sh
@@ -40,11 +40,11 @@ npu_status_pid=$!
 if [ "${gpu_cnt}"="1" ]; then
    ASCEND_RT_VISIBLE_DEVICES=0 llamafactory-cli train ${output_dir}/${run_name}.yaml | tee "${output_dir}/log.txt" &
    train_pid=$!
-   echo "Start train"
+   echo "Start single npu train"
 else
    FORCE_TORCHRUN=1 llamafactory-cli train ${output_dir}/${run_name}.yaml | tee "${output_dir}/log.txt" &
    train_pid=$!
-   echo "Start train"
+   echo "Start multi npu train"
 fi
 
 wait $train_pid