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 <noreply@anthropic.com>

1 files changed, 65 insertions, 0 deletions

diff --git a/collaborativeagents/scripts/run_fp8.sh b/collaborativeagents/scripts/run_fp8.sh
new file mode 100644
index 0000000..54537fa
--- /dev/null
+++ b/collaborativeagents/scripts/run_fp8.sh
@@ -0,0 +1,65 @@
+# vllm serve meta-llama/Llama-3.3-70B-Instruct --port 8004 --tensor-parallel-size 4 --max-model-len 16384 --gpu-memory-utilization 0.9 --quantization fp8 --enforce-eager
+# python -m sglang.launch_server --model-path meta-llama/Llama-3.3-70B-Instruct --port 8004 --tp-size 4 --context-length 16384 --mem-fraction-static 0.9 --quantization fp8
+
+
+BATCH_SIZE=1
+# BATCH_SIZE=20
+
+# Loop over eval sizes and datasets
+for EVAL_SIZE in 5; do
+    for DATASET in logiqa; do #  mmlu medqa humaneval bigcodebench math-500 math-hard
+        # Convert dataset name for file paths (replace - with _)
+        DATASET_FILE=$(echo ${DATASET} | tr '-' '_')
+
+        echo "Running experiments for dataset: ${DATASET} with eval_size ${EVAL_SIZE}"
+
+        # # no_user experiment
+        # python3 run.py --experiment_type no_user --dataset ${DATASET} --eval_size ${EVAL_SIZE} --batch_size ${BATCH_SIZE} \
+        #     --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+        #     --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+        #     --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/no_user/${DATASET_FILE}_llama70b_user_llama70b_agent_no_user_eval_size_${EVAL_SIZE}.jsonl \
+        #     >> ./runs/llama70b_fp8/no_user/${DATASET_FILE}_llama70b_user_llama70b_agent_no_user_eval_size_${EVAL_SIZE}.out 2>&1
+
+        # # user_no_profile experiment
+        # python3 run.py --experiment_type user_no_profile --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \
+        #     --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \
+        #     --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+        #     --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+        #     --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/user_no_profile/${DATASET_FILE}_llama70b_user_llama70b_agent_user_no_profile_eval_size_${EVAL_SIZE}.jsonl \
+        #     >> ./runs/llama70b_fp8/user_no_profile/${DATASET_FILE}_llama70b_user_llama70b_agent_user_no_profile_eval_size_${EVAL_SIZE}.out 2>&1
+
+        # # user_profiles_without_preferences experiment
+        # python3 run.py --experiment_type user_profiles_without_preferences --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \
+        #     --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \
+        #     --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+        #     --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+        #     --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/user_profiles_without_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_user_profiles_without_preferences_eval_size_${EVAL_SIZE}.jsonl \
+        #     >> ./runs/llama70b_fp8/user_profiles_without_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_user_profiles_without_preferences_eval_size_${EVAL_SIZE}.out 2>&1
+
+        # # user_profiles_with_preferences experiment
+        # python3 run.py --experiment_type user_profiles_with_preferences --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \
+        #     --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \
+        #     --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+        #     --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+        #     --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/user_profiles_with_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_user_profiles_with_preferences_eval_size_${EVAL_SIZE}.jsonl \
+        #     >> ./runs/llama70b_fp8/user_profiles_with_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_user_profiles_with_preferences_eval_size_${EVAL_SIZE}.out 2>&1
+
+        # # agent_with_user_preferences experiment
+        # python3 run.py --experiment_type agent_with_user_preferences --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \
+        #     --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \
+        #     --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+        #     --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+        #     --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/agent_with_user_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_agent_with_user_preferences_eval_size_${EVAL_SIZE}_v2.jsonl \
+        #     >> ./runs/llama70b_fp8/agent_with_user_preferences/${DATASET_FILE}_llama70b_user_llama70b_agent_agent_with_user_preferences_eval_size_${EVAL_SIZE}_v2.out 2>&1
+
+        # agent_with_reflection experiment
+        python3 run.py --experiment_type agent_with_reflection --dataset ${DATASET} --eval_size ${EVAL_SIZE} --max_turns 10 --batch_size ${BATCH_SIZE} \
+            --user_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --user_api_base http://localhost:8004/v1 --user_api_key EMPTY \
+            --collaborator_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --collaborator_api_base http://localhost:8004/v1 --collaborator_api_key EMPTY \
+            --judge_model_name hosted_vllm/meta-llama/Llama-3.3-70B-Instruct --judge_api_base http://localhost:8004/v1 --judge_api_key EMPTY \
+            --output_file /shared/storage-01/users/mehri2/mem/collaborativeagents/scripts/runs/llama70b_fp8/agent_with_reflection/${DATASET_FILE}_llama70b_user_llama70b_agent_agent_with_reflection_eval_size_${EVAL_SIZE}.jsonl \
+            >> ./runs/llama70b_fp8/agent_with_reflection/${DATASET_FILE}_llama70b_user_llama70b_agent_agent_with_reflection_eval_size_${EVAL_SIZE}.out 2>&1
+
+
+    done
+done
\ No newline at end of file