From dc801c07cf38b0c495686463e6ca6f871a64440e Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Tue, 27 Jan 2026 09:57:37 -0600 Subject: Add collaborativeagents module and update gitignore - Add collaborativeagents subproject with adapters, agents, and evaluation modules - Update .gitignore to exclude large binary files (.whl, .tar), wandb logs, and results Co-Authored-By: Claude Opus 4.5 --- .../scripts/generate_training_data.sh | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) create mode 100644 collaborativeagents/scripts/generate_training_data.sh (limited to 'collaborativeagents/scripts/generate_training_data.sh') diff --git a/collaborativeagents/scripts/generate_training_data.sh b/collaborativeagents/scripts/generate_training_data.sh new file mode 100644 index 0000000..bdd5fba --- /dev/null +++ b/collaborativeagents/scripts/generate_training_data.sh @@ -0,0 +1,22 @@ +# python -m sglang.launch_server --model-path meta-llama/Llama-3.3-70B-Instruct --port 8004 --tp-size 4 --context-length 16384 + +BATCH_SIZE=100 + +# Loop over eval sizes and datasets +for EVAL_SIZE in 20; do + for DATASET in math-hard math-500 logiqa mmlu medqa; do + # Convert dataset name for file paths (replace - with _) + DATASET_FILE=$(echo ${DATASET} | tr '-' '_') + + echo "Generating training data for dataset: ${DATASET} with eval_size ${EVAL_SIZE}" + + # training_data_with_user_profiles_with_preferences + python3 run.py --experiment_type training_data_with_user_profiles_with_preferences --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \ + --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \ + --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \ + --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \ + --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/training/training_data/${DATASET_FILE}_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_${EVAL_SIZE}.jsonl \ + >> /shared/storage-01/users/mehri2/mem/collaborativeagents/training/training_data/${DATASET_FILE}_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_${EVAL_SIZE}.out 2>&1 + + done +done \ No newline at end of file -- cgit v1.2.3