Merge branch 'main' into kto

applications/ColossalChat/examples/training_scripts/train_dpo.sh

@@ -15,9 +15,8 @@ set_n_least_used_CUDA_VISIBLE_DEVICES() {
 }
 set_n_least_used_CUDA_VISIBLE_DEVICES 4
 
-PROJECT_NAME="dpo"
+PROJECT_NAME="DPO"
 PARENT_SAVE_DIR="" # Path to a folder to save checkpoints
-PARENT_TENSORBOARD_DIR="" # Path to a folder to save logs
 PARENT_CONFIG_FILE="" # Path to a folder to save training config logs
 PRETRAINED_MODEL_PATH="" # huggingface or local model path
 PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
@@ -38,11 +37,10 @@ declare -a dataset=(
 TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
 FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
 SAVE_DIR="${PARENT_SAVE_DIR}${FULL_PROJECT_NAME}"
-CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
+CONFIG_FILE="${PARENT_CONFIG_FILE}${FULL_PROJECT_NAME}.json"
 
 colossalai run --nproc_per_node 4 --hostfile hostfile --master_port 31313 train_dpo.py \
     --pretrain $PRETRAINED_MODEL_PATH \
-    --checkpoint_path $PRETRAINED_MODEL_PATH \
     --tokenizer_dir $PRETRAINED_TOKENIZER_PATH \
     --dataset ${dataset[@]} \
     --plugin "zero2" \

applications/ColossalChat/examples/training_scripts/train_orpo.sh

@@ -13,11 +13,10 @@ set_n_least_used_CUDA_VISIBLE_DEVICES() {
     echo "Now CUDA_VISIBLE_DEVICES is set to:"
     echo "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"
 }
-set_n_least_used_CUDA_VISIBLE_DEVICES 8
+set_n_least_used_CUDA_VISIBLE_DEVICES 2
 
-PROJECT_NAME="dpo"
+PROJECT_NAME="ORPO"
 PARENT_SAVE_DIR="" # Path to a folder to save checkpoints
-PARENT_TENSORBOARD_DIR="" # Path to a folder to save logs
 PARENT_CONFIG_FILE="" # Path to a folder to save training config logs
 PRETRAINED_MODEL_PATH="" # huggingface or local model path
 PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
@@ -38,11 +37,10 @@ declare -a dataset=(
 TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
 FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
 SAVE_DIR="${PARENT_SAVE_DIR}${FULL_PROJECT_NAME}"
-CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
+CONFIG_FILE="${PARENT_CONFIG_FILE}${FULL_PROJECT_NAME}.json"
 
 colossalai run --nproc_per_node 8 --hostfile hostfile --master_port 31313 train_orpo.py \
     --pretrain $PRETRAINED_MODEL_PATH \
-    --checkpoint_path $PRETRAINED_MODEL_PATH \
     --tokenizer_dir $PRETRAINED_TOKENIZER_PATH \
     --dataset ${dataset[@]} \
     --plugin "zero2" \

applications/ColossalChat/examples/training_scripts/train_ppo.sh

@@ -15,10 +15,9 @@ set_n_least_used_CUDA_VISIBLE_DEVICES() {
 }
 set_n_least_used_CUDA_VISIBLE_DEVICES 8
 
-PROJECT_NAME="ppo"
+PROJECT_NAME="PPO"
 
 PARENT_SAVE_DIR="" # Path to a folder to save checkpoints
-PARENT_TENSORBOARD_DIR="" # Path to a folder to save logs
 PARENT_CONFIG_FILE="" # Path to a folder to save training config logs
 PRETRAINED_MODEL_PATH="" # local pretrained model path (from RLHF step 1: SFT)
 PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
@@ -54,7 +53,7 @@ declare -a ptx_dataset=(
 TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
 FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
 SAVE_DIR="${PARENT_SAVE_DIR}${FULL_PROJECT_NAME}"
-CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
+CONFIG_FILE="${PARENT_CONFIG_FILE}${FULL_PROJECT_NAME}.json"
 
 colossalai run --nproc_per_node 8 --hostfile hostfile --master_port 31312 train_ppo.py \
     --pretrain $PRETRAINED_MODEL_PATH \

applications/ColossalChat/examples/training_scripts/train_rm.sh

@@ -15,9 +15,8 @@ set_n_least_used_CUDA_VISIBLE_DEVICES() {
 }
 set_n_least_used_CUDA_VISIBLE_DEVICES 8
 
-PROJECT_NAME="rm"
+PROJECT_NAME="RM"
 PARENT_SAVE_DIR="" # Path to a folder to save checkpoints
-PARENT_TENSORBOARD_DIR="" # Path to a folder to save logs
 PARENT_CONFIG_FILE="" # Path to a folder to save training config logs
 PRETRAINED_MODEL_PATH="" # huggingface or local model path
 PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
@@ -38,7 +37,7 @@ declare -a dataset=(
 TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
 FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
 SAVE_DIR="${PARENT_SAVE_DIR}${FULL_PROJECT_NAME}"
-CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
+CONFIG_FILE="${PARENT_CONFIG_FILE}${FULL_PROJECT_NAME}.json"
 
 colossalai run --nproc_per_node 8 --hostfile hostfile --master_port 31312 train_rm.py \
     --pretrain $PRETRAINED_MODEL_PATH \

applications/ColossalChat/examples/training_scripts/train_sft.py

@@ -61,7 +61,7 @@ def train(args):
         Default torch ddp plugin without any acceleration, for
         debugging purpose acceleration, for debugging purpose
         """
-        plugin = TorchDDPPlugin(find_unused_parameters=True)
+        plugin = TorchDDPPlugin(find_unused_parameters=True if args.grad_checkpoint is False else False)
     elif args.plugin == "gemini":
         plugin = GeminiPlugin(
             precision=args.mixed_precision,

applications/ColossalChat/examples/training_scripts/train_sft.sh

@@ -14,9 +14,8 @@ set_n_least_used_CUDA_VISIBLE_DEVICES() {
 }
 
 set_n_least_used_CUDA_VISIBLE_DEVICES 4
-PROJECT_NAME="sft"
+PROJECT_NAME="SFT"
 PARENT_SAVE_DIR="" # Path to a folder to save checkpoints
-PARENT_TENSORBOARD_DIR="" # Path to a folder to save logs
 PARENT_CONFIG_FILE="" # Path to a folder to save training config logs
 PRETRAINED_MODEL_PATH="" # huggingface or local model path
 PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
@@ -36,7 +35,7 @@ declare -a dataset=(
 TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
 FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
 SAVE_DIR="${PARENT_SAVE_DIR}${FULL_PROJECT_NAME}"
-CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
+CONFIG_FILE="${PARENT_CONFIG_FILE}${FULL_PROJECT_NAME}.json"
 
 echo $(which colossalai)
 echo $(which python)