From dc801c07cf38b0c495686463e6ca6f871a64440e Mon Sep 17 00:00:00 2001
From: YurenHao0426 <blackhao0426@gmail.com>
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 <noreply@anthropic.com>
---
 collaborativeagents/slurm/run_rag_vector_3x.sh | 74 ++++++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 collaborativeagents/slurm/run_rag_vector_3x.sh

(limited to 'collaborativeagents/slurm/run_rag_vector_3x.sh')

diff --git a/collaborativeagents/slurm/run_rag_vector_3x.sh b/collaborativeagents/slurm/run_rag_vector_3x.sh
new file mode 100644
index 0000000..7ca2e3e
--- /dev/null
+++ b/collaborativeagents/slurm/run_rag_vector_3x.sh
@@ -0,0 +1,74 @@
+#!/bin/bash
+#SBATCH --job-name=rag_3x
+#SBATCH --account=bfqt-delta-gpu
+#SBATCH --partition=gpuH200x8
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --gres=gpu:h200:4
+#SBATCH --mem=200G
+#SBATCH --time=4:00:00
+#SBATCH --output=rag_3x_%j.out
+#SBATCH --error=rag_3x_%j.err
+
+set -e
+cd /projects/bfqt/users/yurenh2/ml-projects/personalization-user-model
+source /u/yurenh2/miniforge3/etc/profile.d/conda.sh
+conda activate eval
+
+export HF_HOME=/projects/bfqt/users/yurenh2/hf_cache/huggingface
+export PYTHONPATH="${PWD}/src:${PWD}/collaborativeagents:${PYTHONPATH}"
+export NCCL_P2P_DISABLE=1
+
+MODEL_70B="meta-llama/Llama-3.1-70B-Instruct"
+MODEL_8B="/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/models/llama-3.1-8b-instruct"
+
+pkill -f "vllm.entrypoints" 2>/dev/null || true
+sleep 2
+
+# 70B user simulator on GPUs 0,1
+CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
+    --model $MODEL_70B --port 8004 --tensor-parallel-size 2 \
+    --gpu-memory-utilization 0.95 --max-model-len 16384 \
+    --download-dir $HF_HOME --dtype bfloat16 --disable-log-requests &
+
+# 8B agent on GPUs 2,3 with 40% memory (leaving room for embedding + reranker)
+CUDA_VISIBLE_DEVICES=2,3 python -m vllm.entrypoints.openai.api_server \
+    --model $MODEL_8B --port 8003 --tensor-parallel-size 2 \
+    --gpu-memory-utilization 0.40 --max-model-len 16384 \
+    --dtype bfloat16 --disable-log-requests &
+
+for i in $(seq 1 120); do
+    R1=$(curl -s http://localhost:8004/health > /dev/null 2>&1 && echo 1 || echo 0)
+    R2=$(curl -s http://localhost:8003/health > /dev/null 2>&1 && echo 1 || echo 0)
+    [ "$R1" = "1" ] && [ "$R2" = "1" ] && break
+    sleep 3
+done
+
+cd collaborativeagents/scripts
+
+# Run 3 times with different output directories
+for run in 1 2 3; do
+    echo "========================================="
+    echo "Starting Run $run of 3"
+    echo "========================================="
+
+    python run_experiments.py \
+        --methods rag_vector \
+        --datasets math-hard \
+        --n-profiles 5 \
+        --n-sessions 20 \
+        --use-vllm \
+        --vllm-user-url http://localhost:8004/v1 \
+        --vllm-agent-url http://localhost:8003/v1 \
+        --parallel-profiles 5 \
+        --profile-path ../data/complex_profiles_v2/profiles_200.jsonl \
+        --output-dir ../results/rag_vector_run${run}_$(date +%Y%m%d_%H%M%S)
+
+    echo "Run $run complete"
+    echo ""
+done
+
+echo "All 3 runs complete!"
+
+pkill -f "vllm.entrypoints" 2>/dev/null || true
-- 
cgit v1.2.3