fix pr

Author: YeAnbang

.github/workflows/run_chatgpt_examples.yml

@@ -34,14 +34,14 @@ jobs:
           pip uninstall flash-attn
           pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124
 
-      - name: Install Colossal-AI
-        run: |
-          BUILD_EXT=1 pip install --no-cache-dir -v -e .
-
       - name: Install flash-attn
         run: |
           pip install https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
 
+      - name: Install Colossal-AI
+        run: |
+          BUILD_EXT=1 pip install --no-cache-dir -v -e .
+
       - name: Install ChatGPT
         env:
           CFLAGS: "-O1"

applications/ColossalChat/tests/test_train.sh

@@ -31,7 +31,7 @@ MODELS_DIR=$TEMP_DIR/models_config
 # Skip those tests due to CI tests timeout
 MODELS=('llama')
 ADVANCED_PLUGINS=('zero2' 'sp_split_gather' 'sp_ring' 'sp_all_to_all' 'tp_zero2' '3d' 'gemini' 'gemini_auto' 'zero2_cpu' 'pp' 'tp_pp')
-ADVANCED_PLUGINS=('zero2' 'sp_all_to_all' 'tp_zero2' '3d' 'gemini' 'gemini_auto' 'zero2_cpu' 'pp' 'tp_pp')
+ADVANCED_PLUGINS=('zero2' 'sp_all_to_all' '3d' 'gemini' 'gemini_auto' 'zero2_cpu' 'pp' 'tp_pp')
 PLUGINS=('zero2' '3d' 'gemini' 'gemini_auto' 'zero2_cpu')
 LORA_RANK=('0')  # skip to reduce CI execution time, can pass all locally
 LORA_CONFIG_ENABLE="--lora_config $BASE_DIR/examples/training_scripts/lora_config.json"
@@ -332,6 +332,7 @@ GRAD_CKPTS=('--grad_checkpoint')
 for lora_rank in ${LORA_RANK[@]}; do
     for model in ${MODELS[@]}; do
         for plugin in ${ADVANCED_PLUGINS[@]}; do
+            echo "[Test]: Starting $model-$plugin-$lora_rank"
             if [[ " ${SKIPPED_TESTS[*]} " =~ " $model-$plugin-$lora_rank " ]]; then
                 echo "[Test]: Skipped $model-$plugin-$lora_rank"
                 continue
@@ -389,8 +390,8 @@ for lora_rank in ${LORA_RANK[@]}; do
             if [[ $plugin == "sp_ring" ]]; then
                 enable_sequence_parallelism='--enable_sequence_parallelism'
                 sp_mode='ring'
-                tp='1'
-                sp='2'
+                tp='2'
+                sp='1'
                 bs='8'
                 plugin='3d'
             fi