fix script

scripts/algolr.sh

@@ -12,19 +12,19 @@ train_selected_pairs="${project}/data/train/${modelname}-apps-train-selected_pai
 apps="/lustre/S/nanziyuan/datasets/apps/"
 sft="${project}/data/train/${modelname}-sft.jsonl"
 
-ftmodel="${project}/model/algolr"
+ftmodel="${project}/model/qwen25_coder_inst_7b-algolr"
 
-export CUDA_VISIBLE_DEVICES=0,1,2,3
+### export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 
-# python -m codecritic.cli.algolr \
-#        --model ${model} \
-#        --dataset ${trainset} \
-#        --pairinfo ${train_selected_pairs} \
-#        --apps ${apps} \
-#        --output ${sft} \
-#        --level beginner \
-#        --tp 1
+python -m codecritic.cli.algolr \
+       --model ${model} \
+       --dataset ${trainset} \
+       --pairinfo ${train_selected_pairs} \
+       --apps ${apps} \
+       --output ${sft} \
+       --level beginner \
+       --tp 1
 
 
 deepspeed --module \
@@ -34,7 +34,7 @@ openrlhf.cli.train_sft \
    --input_key question \
    --output_key response \
    --apply_chat_template \
-   --train_batch_size 32 \
+   --train_batch_size 256 \
    --micro_train_batch_size 2 \
    --max_samples 500000 \
    --pretrain ${model} \
@@ -49,4 +49,4 @@ openrlhf.cli.train_sft \
    --learning_rate 5e-6 \
    --load_checkpoint \
    --gradient_checkpointing \
-   --use_tensorboard ${ftmodel}/runs
+   --use_tensorboard "${ftmodel}_log"

scripts/train_orm.sh

@@ -2,6 +2,7 @@ set -xe
 
 dataset="/lustre/S/nanziyuan/projects/ccc/data/train/qwen25_coder_inst-apps-train-reward_dataset.jsonl"
 
+project="/lustre/S/nanziyuan/projects/ccc"
 model="/lustre/S/huangdi/open_for_out/models/Qwen2.5-Coder-7B-Instruct/"
 ftmodel="${project}/model/qwen25_coder_inst_7b-orm"